![](/media/cache/group/default_image.jpg.50x50_q85.jpg)
luke_correlate_f32.c@3:3f71950ceb71, 2017-05-12 (annotated)
- Committer:
- elt14lpo
- Date:
- Fri May 12 06:59:55 2017 +0000
- Revision:
- 3:3f71950ceb71
- Parent:
- arm_correlate_f32.c@2:1161fea84522
- Child:
- 4:a89e836d9faf
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
elt14lpo | 2:1161fea84522 | 1 | /* ---------------------------------------------------------------------------- |
elt14lpo | 2:1161fea84522 | 2 | * Copyright (C) 2010-2014 ARM Limited. All rights reserved. |
elt14lpo | 2:1161fea84522 | 3 | * |
elt14lpo | 2:1161fea84522 | 4 | * $Date: 19. March 2015 |
elt14lpo | 2:1161fea84522 | 5 | * $Revision: V.1.4.5 |
elt14lpo | 2:1161fea84522 | 6 | * |
elt14lpo | 2:1161fea84522 | 7 | * Project: CMSIS DSP Library |
elt14lpo | 2:1161fea84522 | 8 | * Title: arm_correlate_f32.c |
elt14lpo | 2:1161fea84522 | 9 | * |
elt14lpo | 2:1161fea84522 | 10 | * Description: Correlation of floating-point sequences. |
elt14lpo | 2:1161fea84522 | 11 | * |
elt14lpo | 2:1161fea84522 | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
elt14lpo | 2:1161fea84522 | 13 | * |
elt14lpo | 2:1161fea84522 | 14 | * Redistribution and use in source and binary forms, with or without |
elt14lpo | 2:1161fea84522 | 15 | * modification, are permitted provided that the following conditions |
elt14lpo | 2:1161fea84522 | 16 | * are met: |
elt14lpo | 2:1161fea84522 | 17 | * - Redistributions of source code must retain the above copyright |
elt14lpo | 2:1161fea84522 | 18 | * notice, this list of conditions and the following disclaimer. |
elt14lpo | 2:1161fea84522 | 19 | * - Redistributions in binary form must reproduce the above copyright |
elt14lpo | 2:1161fea84522 | 20 | * notice, this list of conditions and the following disclaimer in |
elt14lpo | 2:1161fea84522 | 21 | * the documentation and/or other materials provided with the |
elt14lpo | 2:1161fea84522 | 22 | * distribution. |
elt14lpo | 2:1161fea84522 | 23 | * - Neither the name of ARM LIMITED nor the names of its contributors |
elt14lpo | 2:1161fea84522 | 24 | * may be used to endorse or promote products derived from this |
elt14lpo | 2:1161fea84522 | 25 | * software without specific prior written permission. |
elt14lpo | 1:d0884279b41d | 26 | * |
elt14lpo | 2:1161fea84522 | 27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
elt14lpo | 2:1161fea84522 | 28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
elt14lpo | 2:1161fea84522 | 29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
elt14lpo | 2:1161fea84522 | 30 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
elt14lpo | 2:1161fea84522 | 31 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
elt14lpo | 2:1161fea84522 | 32 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
elt14lpo | 2:1161fea84522 | 33 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
elt14lpo | 2:1161fea84522 | 34 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
elt14lpo | 2:1161fea84522 | 35 | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
elt14lpo | 2:1161fea84522 | 36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
elt14lpo | 2:1161fea84522 | 37 | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
elt14lpo | 2:1161fea84522 | 38 | * POSSIBILITY OF SUCH DAMAGE. |
elt14lpo | 1:d0884279b41d | 39 | * -------------------------------------------------------------------------- */ |
elt14lpo | 1:d0884279b41d | 40 | |
elt14lpo | 1:d0884279b41d | 41 | #include "arm_math.h" |
elt14lpo | 3:3f71950ceb71 | 42 | #include "luke_correlate_f32.h" |
elt14lpo | 1:d0884279b41d | 43 | |
elt14lpo | 2:1161fea84522 | 44 | /** |
elt14lpo | 2:1161fea84522 | 45 | * @ingroup groupFilters |
elt14lpo | 1:d0884279b41d | 46 | */ |
elt14lpo | 1:d0884279b41d | 47 | |
elt14lpo | 2:1161fea84522 | 48 | /** |
elt14lpo | 2:1161fea84522 | 49 | * @defgroup Corr Correlation |
elt14lpo | 2:1161fea84522 | 50 | * |
elt14lpo | 2:1161fea84522 | 51 | * Correlation is a mathematical operation that is similar to convolution. |
elt14lpo | 2:1161fea84522 | 52 | * As with convolution, correlation uses two signals to produce a third signal. |
elt14lpo | 2:1161fea84522 | 53 | * The underlying algorithms in correlation and convolution are identical except that one of the inputs is flipped in convolution. |
elt14lpo | 2:1161fea84522 | 54 | * Correlation is commonly used to measure the similarity between two signals. |
elt14lpo | 2:1161fea84522 | 55 | * It has applications in pattern recognition, cryptanalysis, and searching. |
elt14lpo | 2:1161fea84522 | 56 | * The CMSIS library provides correlation functions for Q7, Q15, Q31 and floating-point data types. |
elt14lpo | 2:1161fea84522 | 57 | * Fast versions of the Q15 and Q31 functions are also provided. |
elt14lpo | 2:1161fea84522 | 58 | * |
elt14lpo | 2:1161fea84522 | 59 | * \par Algorithm |
elt14lpo | 2:1161fea84522 | 60 | * Let <code>a[n]</code> and <code>b[n]</code> be sequences of length <code>srcALen</code> and <code>srcBLen</code> samples respectively. |
elt14lpo | 2:1161fea84522 | 61 | * The convolution of the two signals is denoted by |
elt14lpo | 2:1161fea84522 | 62 | * <pre> |
elt14lpo | 2:1161fea84522 | 63 | * c[n] = a[n] * b[n] |
elt14lpo | 2:1161fea84522 | 64 | * </pre> |
elt14lpo | 2:1161fea84522 | 65 | * In correlation, one of the signals is flipped in time |
elt14lpo | 2:1161fea84522 | 66 | * <pre> |
elt14lpo | 2:1161fea84522 | 67 | * c[n] = a[n] * b[-n] |
elt14lpo | 2:1161fea84522 | 68 | * </pre> |
elt14lpo | 2:1161fea84522 | 69 | * |
elt14lpo | 2:1161fea84522 | 70 | * \par |
elt14lpo | 2:1161fea84522 | 71 | * and this is mathematically defined as |
elt14lpo | 2:1161fea84522 | 72 | * \image html CorrelateEquation.gif |
elt14lpo | 2:1161fea84522 | 73 | * \par |
elt14lpo | 2:1161fea84522 | 74 | * The <code>pSrcA</code> points to the first input vector of length <code>srcALen</code> and <code>pSrcB</code> points to the second input vector of length <code>srcBLen</code>. |
elt14lpo | 2:1161fea84522 | 75 | * The result <code>c[n]</code> is of length <code>2 * max(srcALen, srcBLen) - 1</code> and is defined over the interval <code>n=0, 1, 2, ..., (2 * max(srcALen, srcBLen) - 2)</code>. |
elt14lpo | 2:1161fea84522 | 76 | * The output result is written to <code>pDst</code> and the calling function must allocate <code>2 * max(srcALen, srcBLen) - 1</code> words for the result. |
elt14lpo | 2:1161fea84522 | 77 | * |
elt14lpo | 2:1161fea84522 | 78 | * <b>Note</b> |
elt14lpo | 2:1161fea84522 | 79 | * \par |
elt14lpo | 2:1161fea84522 | 80 | * The <code>pDst</code> should be initialized to all zeros before being used. |
elt14lpo | 2:1161fea84522 | 81 | * |
elt14lpo | 2:1161fea84522 | 82 | * <b>Fixed-Point Behavior</b> |
elt14lpo | 2:1161fea84522 | 83 | * \par |
elt14lpo | 2:1161fea84522 | 84 | * Correlation requires summing up a large number of intermediate products. |
elt14lpo | 2:1161fea84522 | 85 | * As such, the Q7, Q15, and Q31 functions run a risk of overflow and saturation. |
elt14lpo | 2:1161fea84522 | 86 | * Refer to the function specific documentation below for further details of the particular algorithm used. |
elt14lpo | 1:d0884279b41d | 87 | * |
elt14lpo | 2:1161fea84522 | 88 | * |
elt14lpo | 2:1161fea84522 | 89 | * <b>Fast Versions</b> |
elt14lpo | 2:1161fea84522 | 90 | * |
elt14lpo | 2:1161fea84522 | 91 | * \par |
elt14lpo | 2:1161fea84522 | 92 | * Fast versions are supported for Q31 and Q15. Cycles for Fast versions are less compared to Q31 and Q15 of correlate and the design requires |
elt14lpo | 2:1161fea84522 | 93 | * the input signals should be scaled down to avoid intermediate overflows. |
elt14lpo | 1:d0884279b41d | 94 | * |
elt14lpo | 2:1161fea84522 | 95 | * |
elt14lpo | 2:1161fea84522 | 96 | * <b>Opt Versions</b> |
elt14lpo | 1:d0884279b41d | 97 | * |
elt14lpo | 2:1161fea84522 | 98 | * \par |
elt14lpo | 2:1161fea84522 | 99 | * Opt versions are supported for Q15 and Q7. Design uses internal scratch buffer for getting good optimisation. |
elt14lpo | 2:1161fea84522 | 100 | * These versions are optimised in cycles and consumes more memory(Scratch memory) compared to Q15 and Q7 versions of correlate |
elt14lpo | 1:d0884279b41d | 101 | */ |
elt14lpo | 1:d0884279b41d | 102 | |
elt14lpo | 2:1161fea84522 | 103 | /** |
elt14lpo | 2:1161fea84522 | 104 | * @addtogroup Corr |
elt14lpo | 2:1161fea84522 | 105 | * @{ |
elt14lpo | 1:d0884279b41d | 106 | */ |
elt14lpo | 2:1161fea84522 | 107 | /** |
elt14lpo | 2:1161fea84522 | 108 | * @brief Correlation of floating-point sequences. |
elt14lpo | 2:1161fea84522 | 109 | * @param[in] *pSrcA points to the first input sequence. |
elt14lpo | 2:1161fea84522 | 110 | * @param[in] srcALen length of the first input sequence. |
elt14lpo | 2:1161fea84522 | 111 | * @param[in] *pSrcB points to the second input sequence. |
elt14lpo | 2:1161fea84522 | 112 | * @param[in] srcBLen length of the second input sequence. |
elt14lpo | 2:1161fea84522 | 113 | * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1. |
elt14lpo | 2:1161fea84522 | 114 | * @return none. |
elt14lpo | 1:d0884279b41d | 115 | */ |
elt14lpo | 1:d0884279b41d | 116 | |
elt14lpo | 3:3f71950ceb71 | 117 | void luke_correlate_f32( |
elt14lpo | 2:1161fea84522 | 118 | float * pSrcA, |
elt14lpo | 2:1161fea84522 | 119 | int srcALen, |
elt14lpo | 2:1161fea84522 | 120 | float * pSrcB, |
elt14lpo | 2:1161fea84522 | 121 | int srcBLen, |
elt14lpo | 2:1161fea84522 | 122 | float * pDst) |
elt14lpo | 2:1161fea84522 | 123 | { |
elt14lpo | 2:1161fea84522 | 124 | |
elt14lpo | 2:1161fea84522 | 125 | |
elt14lpo | 2:1161fea84522 | 126 | #ifndef ARM_MATH_CM0_FAMILY |
elt14lpo | 2:1161fea84522 | 127 | |
elt14lpo | 2:1161fea84522 | 128 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
elt14lpo | 2:1161fea84522 | 129 | |
elt14lpo | 2:1161fea84522 | 130 | float *pIn1; /* inputA pointer */ |
elt14lpo | 2:1161fea84522 | 131 | float *pIn2; /* inputB pointer */ |
elt14lpo | 2:1161fea84522 | 132 | float *pOut = pDst; /* output pointer */ |
elt14lpo | 2:1161fea84522 | 133 | float *px; /* Intermediate inputA pointer */ |
elt14lpo | 2:1161fea84522 | 134 | float *py; /* Intermediate inputB pointer */ |
elt14lpo | 2:1161fea84522 | 135 | float *pSrc1; /* Intermediate pointers */ |
elt14lpo | 2:1161fea84522 | 136 | float sum, acc0, acc1, acc2, acc3; /* Accumulators */ |
elt14lpo | 2:1161fea84522 | 137 | float x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */ |
elt14lpo | 2:1161fea84522 | 138 | int j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counters */ |
elt14lpo | 2:1161fea84522 | 139 | int32_t inc = 1; /* Destination address modifier */ |
elt14lpo | 1:d0884279b41d | 140 | |
elt14lpo | 1:d0884279b41d | 141 | |
elt14lpo | 2:1161fea84522 | 142 | /* The algorithm implementation is based on the lengths of the inputs. */ |
elt14lpo | 2:1161fea84522 | 143 | /* srcB is always made to slide across srcA. */ |
elt14lpo | 2:1161fea84522 | 144 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
elt14lpo | 2:1161fea84522 | 145 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
elt14lpo | 2:1161fea84522 | 146 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
elt14lpo | 2:1161fea84522 | 147 | /* and the destination pointer modifier, inc is set to -1 */ |
elt14lpo | 2:1161fea84522 | 148 | /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */ |
elt14lpo | 2:1161fea84522 | 149 | /* But to improve the performance, |
elt14lpo | 2:1161fea84522 | 150 | * we assume zeroes in the output instead of zero padding either of the the inputs*/ |
elt14lpo | 2:1161fea84522 | 151 | /* If srcALen > srcBLen, |
elt14lpo | 2:1161fea84522 | 152 | * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */ |
elt14lpo | 2:1161fea84522 | 153 | /* If srcALen < srcBLen, |
elt14lpo | 2:1161fea84522 | 154 | * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */ |
elt14lpo | 2:1161fea84522 | 155 | if(srcALen >= srcBLen) |
elt14lpo | 2:1161fea84522 | 156 | { |
elt14lpo | 2:1161fea84522 | 157 | /* Initialization of inputA pointer */ |
elt14lpo | 2:1161fea84522 | 158 | pIn1 = pSrcA; |
elt14lpo | 2:1161fea84522 | 159 | |
elt14lpo | 2:1161fea84522 | 160 | /* Initialization of inputB pointer */ |
elt14lpo | 2:1161fea84522 | 161 | pIn2 = pSrcB; |
elt14lpo | 2:1161fea84522 | 162 | |
elt14lpo | 2:1161fea84522 | 163 | /* Number of output samples is calculated */ |
elt14lpo | 2:1161fea84522 | 164 | outBlockSize = (2u * srcALen) - 1u; |
elt14lpo | 2:1161fea84522 | 165 | |
elt14lpo | 2:1161fea84522 | 166 | /* When srcALen > srcBLen, zero padding has to be done to srcB |
elt14lpo | 2:1161fea84522 | 167 | * to make their lengths equal. |
elt14lpo | 2:1161fea84522 | 168 | * Instead, (outBlockSize - (srcALen + srcBLen - 1)) |
elt14lpo | 2:1161fea84522 | 169 | * number of output samples are made zero */ |
elt14lpo | 2:1161fea84522 | 170 | j = outBlockSize - (srcALen + (srcBLen - 1u)); |
elt14lpo | 1:d0884279b41d | 171 | |
elt14lpo | 2:1161fea84522 | 172 | /* Updating the pointer position to non zero value */ |
elt14lpo | 2:1161fea84522 | 173 | pOut += j; |
elt14lpo | 2:1161fea84522 | 174 | |
elt14lpo | 2:1161fea84522 | 175 | //while(j > 0u) |
elt14lpo | 2:1161fea84522 | 176 | //{ |
elt14lpo | 2:1161fea84522 | 177 | // /* Zero is stored in the destination buffer */ |
elt14lpo | 2:1161fea84522 | 178 | // *pOut++ = 0.0f; |
elt14lpo | 2:1161fea84522 | 179 | |
elt14lpo | 2:1161fea84522 | 180 | // /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 181 | // j--; |
elt14lpo | 2:1161fea84522 | 182 | //} |
elt14lpo | 1:d0884279b41d | 183 | |
elt14lpo | 2:1161fea84522 | 184 | } |
elt14lpo | 2:1161fea84522 | 185 | else |
elt14lpo | 2:1161fea84522 | 186 | { |
elt14lpo | 2:1161fea84522 | 187 | /* Initialization of inputA pointer */ |
elt14lpo | 2:1161fea84522 | 188 | pIn1 = pSrcB; |
elt14lpo | 2:1161fea84522 | 189 | |
elt14lpo | 2:1161fea84522 | 190 | /* Initialization of inputB pointer */ |
elt14lpo | 2:1161fea84522 | 191 | pIn2 = pSrcA; |
elt14lpo | 2:1161fea84522 | 192 | |
elt14lpo | 2:1161fea84522 | 193 | /* srcBLen is always considered as shorter or equal to srcALen */ |
elt14lpo | 2:1161fea84522 | 194 | j = srcBLen; |
elt14lpo | 2:1161fea84522 | 195 | srcBLen = srcALen; |
elt14lpo | 2:1161fea84522 | 196 | srcALen = j; |
elt14lpo | 2:1161fea84522 | 197 | |
elt14lpo | 2:1161fea84522 | 198 | /* CORR(x, y) = Reverse order(CORR(y, x)) */ |
elt14lpo | 2:1161fea84522 | 199 | /* Hence set the destination pointer to point to the last output sample */ |
elt14lpo | 2:1161fea84522 | 200 | pOut = pDst + ((srcALen + srcBLen) - 2u); |
elt14lpo | 2:1161fea84522 | 201 | |
elt14lpo | 2:1161fea84522 | 202 | /* Destination address modifier is set to -1 */ |
elt14lpo | 2:1161fea84522 | 203 | inc = -1; |
elt14lpo | 2:1161fea84522 | 204 | |
elt14lpo | 2:1161fea84522 | 205 | } |
elt14lpo | 1:d0884279b41d | 206 | |
elt14lpo | 2:1161fea84522 | 207 | /* The function is internally |
elt14lpo | 2:1161fea84522 | 208 | * divided into three parts according to the number of multiplications that has to be |
elt14lpo | 2:1161fea84522 | 209 | * taken place between inputA samples and inputB samples. In the first part of the |
elt14lpo | 2:1161fea84522 | 210 | * algorithm, the multiplications increase by one for every iteration. |
elt14lpo | 2:1161fea84522 | 211 | * In the second part of the algorithm, srcBLen number of multiplications are done. |
elt14lpo | 2:1161fea84522 | 212 | * In the third part of the algorithm, the multiplications decrease by one |
elt14lpo | 2:1161fea84522 | 213 | * for every iteration.*/ |
elt14lpo | 2:1161fea84522 | 214 | /* The algorithm is implemented in three stages. |
elt14lpo | 2:1161fea84522 | 215 | * The loop counters of each stage is initiated here. */ |
elt14lpo | 2:1161fea84522 | 216 | blockSize1 = srcBLen - 1u; |
elt14lpo | 2:1161fea84522 | 217 | blockSize2 = srcALen - (srcBLen - 1u); |
elt14lpo | 2:1161fea84522 | 218 | blockSize3 = blockSize1; |
elt14lpo | 2:1161fea84522 | 219 | |
elt14lpo | 2:1161fea84522 | 220 | /* -------------------------- |
elt14lpo | 2:1161fea84522 | 221 | * Initializations of stage1 |
elt14lpo | 2:1161fea84522 | 222 | * -------------------------*/ |
elt14lpo | 1:d0884279b41d | 223 | |
elt14lpo | 2:1161fea84522 | 224 | /* sum = x[0] * y[srcBlen - 1] |
elt14lpo | 2:1161fea84522 | 225 | * sum = x[0] * y[srcBlen-2] + x[1] * y[srcBlen - 1] |
elt14lpo | 2:1161fea84522 | 226 | * .... |
elt14lpo | 2:1161fea84522 | 227 | * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] |
elt14lpo | 2:1161fea84522 | 228 | */ |
elt14lpo | 2:1161fea84522 | 229 | |
elt14lpo | 2:1161fea84522 | 230 | /* In this stage the MAC operations are increased by 1 for every iteration. |
elt14lpo | 2:1161fea84522 | 231 | The count variable holds the number of MAC operations performed */ |
elt14lpo | 2:1161fea84522 | 232 | count = 1u; |
elt14lpo | 1:d0884279b41d | 233 | |
elt14lpo | 2:1161fea84522 | 234 | /* Working pointer of inputA */ |
elt14lpo | 2:1161fea84522 | 235 | px = pIn1; |
elt14lpo | 2:1161fea84522 | 236 | |
elt14lpo | 2:1161fea84522 | 237 | /* Working pointer of inputB */ |
elt14lpo | 2:1161fea84522 | 238 | pSrc1 = pIn2 + (srcBLen - 1u); |
elt14lpo | 2:1161fea84522 | 239 | py = pSrc1; |
elt14lpo | 2:1161fea84522 | 240 | |
elt14lpo | 2:1161fea84522 | 241 | /* ------------------------ |
elt14lpo | 2:1161fea84522 | 242 | * Stage1 process |
elt14lpo | 2:1161fea84522 | 243 | * ----------------------*/ |
elt14lpo | 1:d0884279b41d | 244 | |
elt14lpo | 2:1161fea84522 | 245 | /* The first stage starts here */ |
elt14lpo | 2:1161fea84522 | 246 | while(blockSize1 > 0u) |
elt14lpo | 2:1161fea84522 | 247 | { |
elt14lpo | 2:1161fea84522 | 248 | /* Accumulator is made zero for every iteration */ |
elt14lpo | 2:1161fea84522 | 249 | sum = 0.0f; |
elt14lpo | 1:d0884279b41d | 250 | |
elt14lpo | 2:1161fea84522 | 251 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
elt14lpo | 2:1161fea84522 | 252 | k = count >> 2u; |
elt14lpo | 1:d0884279b41d | 253 | |
elt14lpo | 2:1161fea84522 | 254 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
elt14lpo | 2:1161fea84522 | 255 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
elt14lpo | 2:1161fea84522 | 256 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 257 | { |
elt14lpo | 2:1161fea84522 | 258 | /* x[0] * y[srcBLen - 4] */ |
elt14lpo | 2:1161fea84522 | 259 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 260 | /* x[1] * y[srcBLen - 3] */ |
elt14lpo | 2:1161fea84522 | 261 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 262 | /* x[2] * y[srcBLen - 2] */ |
elt14lpo | 2:1161fea84522 | 263 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 264 | /* x[3] * y[srcBLen - 1] */ |
elt14lpo | 2:1161fea84522 | 265 | sum += *px++ * *py++; |
elt14lpo | 1:d0884279b41d | 266 | |
elt14lpo | 2:1161fea84522 | 267 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 268 | k--; |
elt14lpo | 2:1161fea84522 | 269 | } |
elt14lpo | 2:1161fea84522 | 270 | |
elt14lpo | 2:1161fea84522 | 271 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
elt14lpo | 2:1161fea84522 | 272 | ** No loop unrolling is used. */ |
elt14lpo | 2:1161fea84522 | 273 | k = count % 0x4u; |
elt14lpo | 1:d0884279b41d | 274 | |
elt14lpo | 2:1161fea84522 | 275 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 276 | { |
elt14lpo | 2:1161fea84522 | 277 | /* Perform the multiply-accumulate */ |
elt14lpo | 2:1161fea84522 | 278 | /* x[0] * y[srcBLen - 1] */ |
elt14lpo | 2:1161fea84522 | 279 | sum += *px++ * *py++; |
elt14lpo | 1:d0884279b41d | 280 | |
elt14lpo | 2:1161fea84522 | 281 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 282 | k--; |
elt14lpo | 1:d0884279b41d | 283 | } |
elt14lpo | 1:d0884279b41d | 284 | |
elt14lpo | 2:1161fea84522 | 285 | /* Store the result in the accumulator in the destination buffer. */ |
elt14lpo | 2:1161fea84522 | 286 | *pOut = sum; |
elt14lpo | 2:1161fea84522 | 287 | /* Destination pointer is updated according to the address modifier, inc */ |
elt14lpo | 2:1161fea84522 | 288 | pOut += inc; |
elt14lpo | 1:d0884279b41d | 289 | |
elt14lpo | 2:1161fea84522 | 290 | /* Update the inputA and inputB pointers for next MAC calculation */ |
elt14lpo | 2:1161fea84522 | 291 | py = pSrc1 - count; |
elt14lpo | 1:d0884279b41d | 292 | px = pIn1; |
elt14lpo | 1:d0884279b41d | 293 | |
elt14lpo | 2:1161fea84522 | 294 | /* Increment the MAC count */ |
elt14lpo | 2:1161fea84522 | 295 | count++; |
elt14lpo | 2:1161fea84522 | 296 | |
elt14lpo | 2:1161fea84522 | 297 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 298 | blockSize1--; |
elt14lpo | 2:1161fea84522 | 299 | } |
elt14lpo | 2:1161fea84522 | 300 | |
elt14lpo | 2:1161fea84522 | 301 | /* -------------------------- |
elt14lpo | 2:1161fea84522 | 302 | * Initializations of stage2 |
elt14lpo | 2:1161fea84522 | 303 | * ------------------------*/ |
elt14lpo | 1:d0884279b41d | 304 | |
elt14lpo | 2:1161fea84522 | 305 | /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] |
elt14lpo | 2:1161fea84522 | 306 | * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] |
elt14lpo | 2:1161fea84522 | 307 | * .... |
elt14lpo | 2:1161fea84522 | 308 | * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
elt14lpo | 2:1161fea84522 | 309 | */ |
elt14lpo | 2:1161fea84522 | 310 | |
elt14lpo | 2:1161fea84522 | 311 | /* Working pointer of inputA */ |
elt14lpo | 2:1161fea84522 | 312 | px = pIn1; |
elt14lpo | 1:d0884279b41d | 313 | |
elt14lpo | 2:1161fea84522 | 314 | /* Working pointer of inputB */ |
elt14lpo | 2:1161fea84522 | 315 | py = pIn2; |
elt14lpo | 2:1161fea84522 | 316 | |
elt14lpo | 2:1161fea84522 | 317 | /* count is index by which the pointer pIn1 to be incremented */ |
elt14lpo | 2:1161fea84522 | 318 | count = 0u; |
elt14lpo | 2:1161fea84522 | 319 | |
elt14lpo | 2:1161fea84522 | 320 | /* ------------------- |
elt14lpo | 2:1161fea84522 | 321 | * Stage2 process |
elt14lpo | 2:1161fea84522 | 322 | * ------------------*/ |
elt14lpo | 1:d0884279b41d | 323 | |
elt14lpo | 2:1161fea84522 | 324 | /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. |
elt14lpo | 2:1161fea84522 | 325 | * So, to loop unroll over blockSize2, |
elt14lpo | 2:1161fea84522 | 326 | * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */ |
elt14lpo | 2:1161fea84522 | 327 | if(srcBLen >= 4u) |
elt14lpo | 2:1161fea84522 | 328 | { |
elt14lpo | 2:1161fea84522 | 329 | /* Loop unroll over blockSize2, by 4 */ |
elt14lpo | 2:1161fea84522 | 330 | blkCnt = blockSize2 >> 2u; |
elt14lpo | 2:1161fea84522 | 331 | |
elt14lpo | 2:1161fea84522 | 332 | while(blkCnt > 0u) |
elt14lpo | 2:1161fea84522 | 333 | { |
elt14lpo | 2:1161fea84522 | 334 | /* Set all accumulators to zero */ |
elt14lpo | 2:1161fea84522 | 335 | acc0 = 0.0f; |
elt14lpo | 2:1161fea84522 | 336 | acc1 = 0.0f; |
elt14lpo | 2:1161fea84522 | 337 | acc2 = 0.0f; |
elt14lpo | 2:1161fea84522 | 338 | acc3 = 0.0f; |
elt14lpo | 2:1161fea84522 | 339 | |
elt14lpo | 2:1161fea84522 | 340 | /* read x[0], x[1], x[2] samples */ |
elt14lpo | 2:1161fea84522 | 341 | x0 = *(px++); |
elt14lpo | 2:1161fea84522 | 342 | x1 = *(px++); |
elt14lpo | 2:1161fea84522 | 343 | x2 = *(px++); |
elt14lpo | 1:d0884279b41d | 344 | |
elt14lpo | 2:1161fea84522 | 345 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
elt14lpo | 2:1161fea84522 | 346 | k = srcBLen >> 2u; |
elt14lpo | 2:1161fea84522 | 347 | |
elt14lpo | 2:1161fea84522 | 348 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
elt14lpo | 2:1161fea84522 | 349 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
elt14lpo | 2:1161fea84522 | 350 | do |
elt14lpo | 2:1161fea84522 | 351 | { |
elt14lpo | 2:1161fea84522 | 352 | /* Read y[0] sample */ |
elt14lpo | 2:1161fea84522 | 353 | c0 = *(py++); |
elt14lpo | 2:1161fea84522 | 354 | |
elt14lpo | 2:1161fea84522 | 355 | /* Read x[3] sample */ |
elt14lpo | 2:1161fea84522 | 356 | x3 = *(px++); |
elt14lpo | 2:1161fea84522 | 357 | |
elt14lpo | 2:1161fea84522 | 358 | /* Perform the multiply-accumulate */ |
elt14lpo | 2:1161fea84522 | 359 | /* acc0 += x[0] * y[0] */ |
elt14lpo | 2:1161fea84522 | 360 | acc0 += x0 * c0; |
elt14lpo | 2:1161fea84522 | 361 | /* acc1 += x[1] * y[0] */ |
elt14lpo | 2:1161fea84522 | 362 | acc1 += x1 * c0; |
elt14lpo | 2:1161fea84522 | 363 | /* acc2 += x[2] * y[0] */ |
elt14lpo | 2:1161fea84522 | 364 | acc2 += x2 * c0; |
elt14lpo | 2:1161fea84522 | 365 | /* acc3 += x[3] * y[0] */ |
elt14lpo | 2:1161fea84522 | 366 | acc3 += x3 * c0; |
elt14lpo | 1:d0884279b41d | 367 | |
elt14lpo | 2:1161fea84522 | 368 | /* Read y[1] sample */ |
elt14lpo | 2:1161fea84522 | 369 | c0 = *(py++); |
elt14lpo | 2:1161fea84522 | 370 | |
elt14lpo | 2:1161fea84522 | 371 | /* Read x[4] sample */ |
elt14lpo | 2:1161fea84522 | 372 | x0 = *(px++); |
elt14lpo | 2:1161fea84522 | 373 | |
elt14lpo | 2:1161fea84522 | 374 | /* Perform the multiply-accumulate */ |
elt14lpo | 2:1161fea84522 | 375 | /* acc0 += x[1] * y[1] */ |
elt14lpo | 2:1161fea84522 | 376 | acc0 += x1 * c0; |
elt14lpo | 2:1161fea84522 | 377 | /* acc1 += x[2] * y[1] */ |
elt14lpo | 2:1161fea84522 | 378 | acc1 += x2 * c0; |
elt14lpo | 2:1161fea84522 | 379 | /* acc2 += x[3] * y[1] */ |
elt14lpo | 2:1161fea84522 | 380 | acc2 += x3 * c0; |
elt14lpo | 2:1161fea84522 | 381 | /* acc3 += x[4] * y[1] */ |
elt14lpo | 2:1161fea84522 | 382 | acc3 += x0 * c0; |
elt14lpo | 2:1161fea84522 | 383 | |
elt14lpo | 2:1161fea84522 | 384 | /* Read y[2] sample */ |
elt14lpo | 2:1161fea84522 | 385 | c0 = *(py++); |
elt14lpo | 1:d0884279b41d | 386 | |
elt14lpo | 2:1161fea84522 | 387 | /* Read x[5] sample */ |
elt14lpo | 2:1161fea84522 | 388 | x1 = *(px++); |
elt14lpo | 1:d0884279b41d | 389 | |
elt14lpo | 2:1161fea84522 | 390 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 391 | /* acc0 += x[2] * y[2] */ |
elt14lpo | 2:1161fea84522 | 392 | acc0 += x2 * c0; |
elt14lpo | 2:1161fea84522 | 393 | /* acc1 += x[3] * y[2] */ |
elt14lpo | 2:1161fea84522 | 394 | acc1 += x3 * c0; |
elt14lpo | 2:1161fea84522 | 395 | /* acc2 += x[4] * y[2] */ |
elt14lpo | 2:1161fea84522 | 396 | acc2 += x0 * c0; |
elt14lpo | 2:1161fea84522 | 397 | /* acc3 += x[5] * y[2] */ |
elt14lpo | 2:1161fea84522 | 398 | acc3 += x1 * c0; |
elt14lpo | 2:1161fea84522 | 399 | |
elt14lpo | 2:1161fea84522 | 400 | /* Read y[3] sample */ |
elt14lpo | 2:1161fea84522 | 401 | c0 = *(py++); |
elt14lpo | 2:1161fea84522 | 402 | |
elt14lpo | 2:1161fea84522 | 403 | /* Read x[6] sample */ |
elt14lpo | 2:1161fea84522 | 404 | x2 = *(px++); |
elt14lpo | 1:d0884279b41d | 405 | |
elt14lpo | 2:1161fea84522 | 406 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 407 | /* acc0 += x[3] * y[3] */ |
elt14lpo | 2:1161fea84522 | 408 | acc0 += x3 * c0; |
elt14lpo | 2:1161fea84522 | 409 | /* acc1 += x[4] * y[3] */ |
elt14lpo | 2:1161fea84522 | 410 | acc1 += x0 * c0; |
elt14lpo | 2:1161fea84522 | 411 | /* acc2 += x[5] * y[3] */ |
elt14lpo | 2:1161fea84522 | 412 | acc2 += x1 * c0; |
elt14lpo | 2:1161fea84522 | 413 | /* acc3 += x[6] * y[3] */ |
elt14lpo | 2:1161fea84522 | 414 | acc3 += x2 * c0; |
elt14lpo | 2:1161fea84522 | 415 | |
elt14lpo | 2:1161fea84522 | 416 | |
elt14lpo | 2:1161fea84522 | 417 | } while(--k); |
elt14lpo | 2:1161fea84522 | 418 | |
elt14lpo | 2:1161fea84522 | 419 | /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. |
elt14lpo | 2:1161fea84522 | 420 | ** No loop unrolling is used. */ |
elt14lpo | 2:1161fea84522 | 421 | k = srcBLen % 0x4u; |
elt14lpo | 1:d0884279b41d | 422 | |
elt14lpo | 2:1161fea84522 | 423 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 424 | { |
elt14lpo | 2:1161fea84522 | 425 | /* Read y[4] sample */ |
elt14lpo | 2:1161fea84522 | 426 | c0 = *(py++); |
elt14lpo | 2:1161fea84522 | 427 | |
elt14lpo | 2:1161fea84522 | 428 | /* Read x[7] sample */ |
elt14lpo | 2:1161fea84522 | 429 | x3 = *(px++); |
elt14lpo | 1:d0884279b41d | 430 | |
elt14lpo | 2:1161fea84522 | 431 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 432 | /* acc0 += x[4] * y[4] */ |
elt14lpo | 2:1161fea84522 | 433 | acc0 += x0 * c0; |
elt14lpo | 2:1161fea84522 | 434 | /* acc1 += x[5] * y[4] */ |
elt14lpo | 2:1161fea84522 | 435 | acc1 += x1 * c0; |
elt14lpo | 2:1161fea84522 | 436 | /* acc2 += x[6] * y[4] */ |
elt14lpo | 2:1161fea84522 | 437 | acc2 += x2 * c0; |
elt14lpo | 2:1161fea84522 | 438 | /* acc3 += x[7] * y[4] */ |
elt14lpo | 2:1161fea84522 | 439 | acc3 += x3 * c0; |
elt14lpo | 1:d0884279b41d | 440 | |
elt14lpo | 2:1161fea84522 | 441 | /* Reuse the present samples for the next MAC */ |
elt14lpo | 2:1161fea84522 | 442 | x0 = x1; |
elt14lpo | 2:1161fea84522 | 443 | x1 = x2; |
elt14lpo | 2:1161fea84522 | 444 | x2 = x3; |
elt14lpo | 1:d0884279b41d | 445 | |
elt14lpo | 1:d0884279b41d | 446 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 447 | k--; |
elt14lpo | 2:1161fea84522 | 448 | } |
elt14lpo | 2:1161fea84522 | 449 | |
elt14lpo | 2:1161fea84522 | 450 | /* Store the result in the accumulator in the destination buffer. */ |
elt14lpo | 2:1161fea84522 | 451 | *pOut = acc0; |
elt14lpo | 2:1161fea84522 | 452 | /* Destination pointer is updated according to the address modifier, inc */ |
elt14lpo | 2:1161fea84522 | 453 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 454 | |
elt14lpo | 2:1161fea84522 | 455 | *pOut = acc1; |
elt14lpo | 2:1161fea84522 | 456 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 457 | |
elt14lpo | 2:1161fea84522 | 458 | *pOut = acc2; |
elt14lpo | 2:1161fea84522 | 459 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 460 | |
elt14lpo | 2:1161fea84522 | 461 | *pOut = acc3; |
elt14lpo | 2:1161fea84522 | 462 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 463 | |
elt14lpo | 2:1161fea84522 | 464 | /* Increment the pointer pIn1 index, count by 4 */ |
elt14lpo | 2:1161fea84522 | 465 | count += 4u; |
elt14lpo | 2:1161fea84522 | 466 | |
elt14lpo | 2:1161fea84522 | 467 | /* Update the inputA and inputB pointers for next MAC calculation */ |
elt14lpo | 2:1161fea84522 | 468 | px = pIn1 + count; |
elt14lpo | 2:1161fea84522 | 469 | py = pIn2; |
elt14lpo | 2:1161fea84522 | 470 | |
elt14lpo | 2:1161fea84522 | 471 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 472 | blkCnt--; |
elt14lpo | 1:d0884279b41d | 473 | } |
elt14lpo | 1:d0884279b41d | 474 | |
elt14lpo | 2:1161fea84522 | 475 | /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. |
elt14lpo | 2:1161fea84522 | 476 | ** No loop unrolling is used. */ |
elt14lpo | 2:1161fea84522 | 477 | blkCnt = blockSize2 % 0x4u; |
elt14lpo | 2:1161fea84522 | 478 | |
elt14lpo | 2:1161fea84522 | 479 | while(blkCnt > 0u) |
elt14lpo | 2:1161fea84522 | 480 | { |
elt14lpo | 2:1161fea84522 | 481 | /* Accumulator is made zero for every iteration */ |
elt14lpo | 2:1161fea84522 | 482 | sum = 0.0f; |
elt14lpo | 2:1161fea84522 | 483 | |
elt14lpo | 2:1161fea84522 | 484 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
elt14lpo | 2:1161fea84522 | 485 | k = srcBLen >> 2u; |
elt14lpo | 2:1161fea84522 | 486 | |
elt14lpo | 2:1161fea84522 | 487 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
elt14lpo | 2:1161fea84522 | 488 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
elt14lpo | 2:1161fea84522 | 489 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 490 | { |
elt14lpo | 2:1161fea84522 | 491 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 492 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 493 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 494 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 495 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 496 | |
elt14lpo | 2:1161fea84522 | 497 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 498 | k--; |
elt14lpo | 2:1161fea84522 | 499 | } |
elt14lpo | 2:1161fea84522 | 500 | |
elt14lpo | 2:1161fea84522 | 501 | /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. |
elt14lpo | 2:1161fea84522 | 502 | ** No loop unrolling is used. */ |
elt14lpo | 2:1161fea84522 | 503 | k = srcBLen % 0x4u; |
elt14lpo | 2:1161fea84522 | 504 | |
elt14lpo | 2:1161fea84522 | 505 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 506 | { |
elt14lpo | 2:1161fea84522 | 507 | /* Perform the multiply-accumulate */ |
elt14lpo | 2:1161fea84522 | 508 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 509 | |
elt14lpo | 2:1161fea84522 | 510 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 511 | k--; |
elt14lpo | 2:1161fea84522 | 512 | } |
elt14lpo | 2:1161fea84522 | 513 | |
elt14lpo | 2:1161fea84522 | 514 | /* Store the result in the accumulator in the destination buffer. */ |
elt14lpo | 2:1161fea84522 | 515 | *pOut = sum; |
elt14lpo | 2:1161fea84522 | 516 | /* Destination pointer is updated according to the address modifier, inc */ |
elt14lpo | 2:1161fea84522 | 517 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 518 | |
elt14lpo | 2:1161fea84522 | 519 | /* Increment the pointer pIn1 index, count by 1 */ |
elt14lpo | 2:1161fea84522 | 520 | count++; |
elt14lpo | 2:1161fea84522 | 521 | |
elt14lpo | 2:1161fea84522 | 522 | /* Update the inputA and inputB pointers for next MAC calculation */ |
elt14lpo | 2:1161fea84522 | 523 | px = pIn1 + count; |
elt14lpo | 2:1161fea84522 | 524 | py = pIn2; |
elt14lpo | 2:1161fea84522 | 525 | |
elt14lpo | 2:1161fea84522 | 526 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 527 | blkCnt--; |
elt14lpo | 2:1161fea84522 | 528 | } |
elt14lpo | 2:1161fea84522 | 529 | } |
elt14lpo | 2:1161fea84522 | 530 | else |
elt14lpo | 2:1161fea84522 | 531 | { |
elt14lpo | 2:1161fea84522 | 532 | /* If the srcBLen is not a multiple of 4, |
elt14lpo | 2:1161fea84522 | 533 | * the blockSize2 loop cannot be unrolled by 4 */ |
elt14lpo | 2:1161fea84522 | 534 | blkCnt = blockSize2; |
elt14lpo | 2:1161fea84522 | 535 | |
elt14lpo | 2:1161fea84522 | 536 | while(blkCnt > 0u) |
elt14lpo | 2:1161fea84522 | 537 | { |
elt14lpo | 2:1161fea84522 | 538 | /* Accumulator is made zero for every iteration */ |
elt14lpo | 2:1161fea84522 | 539 | sum = 0.0f; |
elt14lpo | 2:1161fea84522 | 540 | |
elt14lpo | 2:1161fea84522 | 541 | /* Loop over srcBLen */ |
elt14lpo | 2:1161fea84522 | 542 | k = srcBLen; |
elt14lpo | 2:1161fea84522 | 543 | |
elt14lpo | 2:1161fea84522 | 544 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 545 | { |
elt14lpo | 2:1161fea84522 | 546 | /* Perform the multiply-accumulate */ |
elt14lpo | 2:1161fea84522 | 547 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 548 | |
elt14lpo | 2:1161fea84522 | 549 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 550 | k--; |
elt14lpo | 2:1161fea84522 | 551 | } |
elt14lpo | 2:1161fea84522 | 552 | |
elt14lpo | 2:1161fea84522 | 553 | /* Store the result in the accumulator in the destination buffer. */ |
elt14lpo | 2:1161fea84522 | 554 | *pOut = sum; |
elt14lpo | 2:1161fea84522 | 555 | /* Destination pointer is updated according to the address modifier, inc */ |
elt14lpo | 2:1161fea84522 | 556 | pOut += inc; |
elt14lpo | 1:d0884279b41d | 557 | |
elt14lpo | 2:1161fea84522 | 558 | /* Increment the pointer pIn1 index, count by 1 */ |
elt14lpo | 2:1161fea84522 | 559 | count++; |
elt14lpo | 2:1161fea84522 | 560 | |
elt14lpo | 2:1161fea84522 | 561 | /* Update the inputA and inputB pointers for next MAC calculation */ |
elt14lpo | 2:1161fea84522 | 562 | px = pIn1 + count; |
elt14lpo | 2:1161fea84522 | 563 | py = pIn2; |
elt14lpo | 2:1161fea84522 | 564 | |
elt14lpo | 2:1161fea84522 | 565 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 566 | blkCnt--; |
elt14lpo | 2:1161fea84522 | 567 | } |
elt14lpo | 2:1161fea84522 | 568 | } |
elt14lpo | 2:1161fea84522 | 569 | |
elt14lpo | 2:1161fea84522 | 570 | /* -------------------------- |
elt14lpo | 2:1161fea84522 | 571 | * Initializations of stage3 |
elt14lpo | 2:1161fea84522 | 572 | * -------------------------*/ |
elt14lpo | 2:1161fea84522 | 573 | |
elt14lpo | 2:1161fea84522 | 574 | /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
elt14lpo | 2:1161fea84522 | 575 | * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
elt14lpo | 2:1161fea84522 | 576 | * .... |
elt14lpo | 2:1161fea84522 | 577 | * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] |
elt14lpo | 2:1161fea84522 | 578 | * sum += x[srcALen-1] * y[0] |
elt14lpo | 2:1161fea84522 | 579 | */ |
elt14lpo | 2:1161fea84522 | 580 | |
elt14lpo | 2:1161fea84522 | 581 | /* In this stage the MAC operations are decreased by 1 for every iteration. |
elt14lpo | 2:1161fea84522 | 582 | The count variable holds the number of MAC operations performed */ |
elt14lpo | 2:1161fea84522 | 583 | count = srcBLen - 1u; |
elt14lpo | 2:1161fea84522 | 584 | |
elt14lpo | 2:1161fea84522 | 585 | /* Working pointer of inputA */ |
elt14lpo | 2:1161fea84522 | 586 | pSrc1 = pIn1 + (srcALen - (srcBLen - 1u)); |
elt14lpo | 2:1161fea84522 | 587 | px = pSrc1; |
elt14lpo | 2:1161fea84522 | 588 | |
elt14lpo | 2:1161fea84522 | 589 | /* Working pointer of inputB */ |
elt14lpo | 2:1161fea84522 | 590 | py = pIn2; |
elt14lpo | 2:1161fea84522 | 591 | |
elt14lpo | 2:1161fea84522 | 592 | /* ------------------- |
elt14lpo | 2:1161fea84522 | 593 | * Stage3 process |
elt14lpo | 2:1161fea84522 | 594 | * ------------------*/ |
elt14lpo | 1:d0884279b41d | 595 | |
elt14lpo | 2:1161fea84522 | 596 | while(blockSize3 > 0u) |
elt14lpo | 2:1161fea84522 | 597 | { |
elt14lpo | 2:1161fea84522 | 598 | /* Accumulator is made zero for every iteration */ |
elt14lpo | 2:1161fea84522 | 599 | sum = 0.0f; |
elt14lpo | 2:1161fea84522 | 600 | |
elt14lpo | 2:1161fea84522 | 601 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
elt14lpo | 2:1161fea84522 | 602 | k = count >> 2u; |
elt14lpo | 2:1161fea84522 | 603 | |
elt14lpo | 2:1161fea84522 | 604 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
elt14lpo | 2:1161fea84522 | 605 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
elt14lpo | 2:1161fea84522 | 606 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 607 | { |
elt14lpo | 2:1161fea84522 | 608 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 609 | /* sum += x[srcALen - srcBLen + 4] * y[3] */ |
elt14lpo | 2:1161fea84522 | 610 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 611 | /* sum += x[srcALen - srcBLen + 3] * y[2] */ |
elt14lpo | 2:1161fea84522 | 612 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 613 | /* sum += x[srcALen - srcBLen + 2] * y[1] */ |
elt14lpo | 2:1161fea84522 | 614 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 615 | /* sum += x[srcALen - srcBLen + 1] * y[0] */ |
elt14lpo | 2:1161fea84522 | 616 | sum += *px++ * *py++; |
elt14lpo | 1:d0884279b41d | 617 | |
elt14lpo | 2:1161fea84522 | 618 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 619 | k--; |
elt14lpo | 2:1161fea84522 | 620 | } |
elt14lpo | 2:1161fea84522 | 621 | |
elt14lpo | 2:1161fea84522 | 622 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
elt14lpo | 2:1161fea84522 | 623 | ** No loop unrolling is used. */ |
elt14lpo | 2:1161fea84522 | 624 | k = count % 0x4u; |
elt14lpo | 2:1161fea84522 | 625 | |
elt14lpo | 2:1161fea84522 | 626 | while(k > 0u) |
elt14lpo | 2:1161fea84522 | 627 | { |
elt14lpo | 2:1161fea84522 | 628 | /* Perform the multiply-accumulates */ |
elt14lpo | 2:1161fea84522 | 629 | sum += *px++ * *py++; |
elt14lpo | 2:1161fea84522 | 630 | |
elt14lpo | 2:1161fea84522 | 631 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 632 | k--; |
elt14lpo | 2:1161fea84522 | 633 | } |
elt14lpo | 2:1161fea84522 | 634 | |
elt14lpo | 2:1161fea84522 | 635 | /* Store the result in the accumulator in the destination buffer. */ |
elt14lpo | 2:1161fea84522 | 636 | *pOut = sum; |
elt14lpo | 2:1161fea84522 | 637 | /* Destination pointer is updated according to the address modifier, inc */ |
elt14lpo | 2:1161fea84522 | 638 | pOut += inc; |
elt14lpo | 2:1161fea84522 | 639 | |
elt14lpo | 2:1161fea84522 | 640 | /* Update the inputA and inputB pointers for next MAC calculation */ |
elt14lpo | 2:1161fea84522 | 641 | px = ++pSrc1; |
elt14lpo | 1:d0884279b41d | 642 | py = pIn2; |
elt14lpo | 1:d0884279b41d | 643 | |
elt14lpo | 2:1161fea84522 | 644 | /* Decrement the MAC count */ |
elt14lpo | 2:1161fea84522 | 645 | count--; |
elt14lpo | 1:d0884279b41d | 646 | |
elt14lpo | 2:1161fea84522 | 647 | /* Decrement the loop counter */ |
elt14lpo | 2:1161fea84522 | 648 | blockSize3--; |
elt14lpo | 2:1161fea84522 | 649 | } |
elt14lpo | 1:d0884279b41d | 650 | |
elt14lpo | 2:1161fea84522 | 651 | #else |
elt14lpo | 1:d0884279b41d | 652 | |
elt14lpo | 2:1161fea84522 | 653 | /* Run the below code for Cortex-M0 */ |
elt14lpo | 1:d0884279b41d | 654 | |
elt14lpo | 2:1161fea84522 | 655 | float *pIn1 = pSrcA; /* inputA pointer */ |
elt14lpo | 2:1161fea84522 | 656 | float *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */ |
elt14lpo | 2:1161fea84522 | 657 | float sum; /* Accumulator */ |
elt14lpo | 2:1161fea84522 | 658 | int i = 0u, j; /* loop counters */ |
elt14lpo | 2:1161fea84522 | 659 | int inv = 0u; /* Reverse order flag */ |
elt14lpo | 2:1161fea84522 | 660 | int tot = 0u; /* Length */ |
elt14lpo | 1:d0884279b41d | 661 | |
elt14lpo | 2:1161fea84522 | 662 | /* The algorithm implementation is based on the lengths of the inputs. */ |
elt14lpo | 2:1161fea84522 | 663 | /* srcB is always made to slide across srcA. */ |
elt14lpo | 2:1161fea84522 | 664 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
elt14lpo | 2:1161fea84522 | 665 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
elt14lpo | 2:1161fea84522 | 666 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
elt14lpo | 2:1161fea84522 | 667 | /* and a varaible, inv is set to 1 */ |
elt14lpo | 2:1161fea84522 | 668 | /* If lengths are not equal then zero pad has to be done to make the two |
elt14lpo | 2:1161fea84522 | 669 | * inputs of same length. But to improve the performance, we assume zeroes |
elt14lpo | 2:1161fea84522 | 670 | * in the output instead of zero padding either of the the inputs*/ |
elt14lpo | 2:1161fea84522 | 671 | /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the |
elt14lpo | 2:1161fea84522 | 672 | * starting of the output buffer */ |
elt14lpo | 2:1161fea84522 | 673 | /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the |
elt14lpo | 2:1161fea84522 | 674 | * ending of the output buffer */ |
elt14lpo | 2:1161fea84522 | 675 | /* Once the zero padding is done the remaining of the output is calcualted |
elt14lpo | 2:1161fea84522 | 676 | * using convolution but with the shorter signal time shifted. */ |
elt14lpo | 1:d0884279b41d | 677 | |
elt14lpo | 2:1161fea84522 | 678 | /* Calculate the length of the remaining sequence */ |
elt14lpo | 2:1161fea84522 | 679 | tot = ((srcALen + srcBLen) - 2u); |
elt14lpo | 1:d0884279b41d | 680 | |
elt14lpo | 2:1161fea84522 | 681 | if(srcALen > srcBLen) |
elt14lpo | 2:1161fea84522 | 682 | { |
elt14lpo | 2:1161fea84522 | 683 | /* Calculating the number of zeros to be padded to the output */ |
elt14lpo | 2:1161fea84522 | 684 | j = srcALen - srcBLen; |
elt14lpo | 1:d0884279b41d | 685 | |
elt14lpo | 2:1161fea84522 | 686 | /* Initialise the pointer after zero padding */ |
elt14lpo | 2:1161fea84522 | 687 | pDst += j; |
elt14lpo | 2:1161fea84522 | 688 | } |
elt14lpo | 1:d0884279b41d | 689 | |
elt14lpo | 2:1161fea84522 | 690 | else if(srcALen < srcBLen) |
elt14lpo | 2:1161fea84522 | 691 | { |
elt14lpo | 2:1161fea84522 | 692 | /* Initialization to inputB pointer */ |
elt14lpo | 2:1161fea84522 | 693 | pIn1 = pSrcB; |
elt14lpo | 1:d0884279b41d | 694 | |
elt14lpo | 2:1161fea84522 | 695 | /* Initialization to the end of inputA pointer */ |
elt14lpo | 2:1161fea84522 | 696 | pIn2 = pSrcA + (srcALen - 1u); |
elt14lpo | 1:d0884279b41d | 697 | |
elt14lpo | 2:1161fea84522 | 698 | /* Initialisation of the pointer after zero padding */ |
elt14lpo | 2:1161fea84522 | 699 | pDst = pDst + tot; |
elt14lpo | 1:d0884279b41d | 700 | |
elt14lpo | 2:1161fea84522 | 701 | /* Swapping the lengths */ |
elt14lpo | 2:1161fea84522 | 702 | j = srcALen; |
elt14lpo | 2:1161fea84522 | 703 | srcALen = srcBLen; |
elt14lpo | 2:1161fea84522 | 704 | srcBLen = j; |
elt14lpo | 1:d0884279b41d | 705 | |
elt14lpo | 2:1161fea84522 | 706 | /* Setting the reverse flag */ |
elt14lpo | 2:1161fea84522 | 707 | inv = 1; |
elt14lpo | 1:d0884279b41d | 708 | |
elt14lpo | 2:1161fea84522 | 709 | } |
elt14lpo | 1:d0884279b41d | 710 | |
elt14lpo | 2:1161fea84522 | 711 | /* Loop to calculate convolution for output length number of times */ |
elt14lpo | 2:1161fea84522 | 712 | for (i = 0u; i <= tot; i++) |
elt14lpo | 2:1161fea84522 | 713 | { |
elt14lpo | 2:1161fea84522 | 714 | /* Initialize sum with zero to carry on MAC operations */ |
elt14lpo | 2:1161fea84522 | 715 | sum = 0.0f; |
elt14lpo | 1:d0884279b41d | 716 | |
elt14lpo | 2:1161fea84522 | 717 | /* Loop to perform MAC operations according to convolution equation */ |
elt14lpo | 2:1161fea84522 | 718 | for (j = 0u; j <= i; j++) |
elt14lpo | 2:1161fea84522 | 719 | { |
elt14lpo | 2:1161fea84522 | 720 | /* Check the array limitations */ |
elt14lpo | 2:1161fea84522 | 721 | if((((i - j) < srcBLen) && (j < srcALen))) |
elt14lpo | 2:1161fea84522 | 722 | { |
elt14lpo | 2:1161fea84522 | 723 | /* z[i] += x[i-j] * y[j] */ |
elt14lpo | 2:1161fea84522 | 724 | sum += pIn1[j] * pIn2[-((int32_t) i - j)]; |
elt14lpo | 2:1161fea84522 | 725 | } |
elt14lpo | 2:1161fea84522 | 726 | } |
elt14lpo | 2:1161fea84522 | 727 | /* Store the output in the destination buffer */ |
elt14lpo | 2:1161fea84522 | 728 | if(inv == 1) |
elt14lpo | 2:1161fea84522 | 729 | *pDst-- = sum; |
elt14lpo | 2:1161fea84522 | 730 | else |
elt14lpo | 2:1161fea84522 | 731 | *pDst++ = sum; |
elt14lpo | 2:1161fea84522 | 732 | } |
elt14lpo | 1:d0884279b41d | 733 | |
elt14lpo | 2:1161fea84522 | 734 | #endif /* #ifndef ARM_MATH_CM0_FAMILY */ |
elt14lpo | 1:d0884279b41d | 735 | |
elt14lpo | 1:d0884279b41d | 736 | } |
elt14lpo | 1:d0884279b41d | 737 | |
elt14lpo | 2:1161fea84522 | 738 | /** |
elt14lpo | 2:1161fea84522 | 739 | * @} end of Corr group |
elt14lpo | 2:1161fea84522 | 740 | */ |