CMSIS DSP library
Dependents: KL25Z_FFT_Demo Hat_Board_v5_1 KL25Z_FFT_Demo_tony KL25Z_FFT_Demo_tony ... more
Fork of mbed-dsp by
cmsis_dsp/FilteringFunctions/arm_correlate_q31.c@1:fdd22bb7aa52, 2012-11-28 (annotated)
- Committer:
- emilmont
- Date:
- Wed Nov 28 12:30:09 2012 +0000
- Revision:
- 1:fdd22bb7aa52
- Child:
- 2:da51fb522205
DSP library code
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emilmont | 1:fdd22bb7aa52 | 1 | /* ---------------------------------------------------------------------- |
emilmont | 1:fdd22bb7aa52 | 2 | * Copyright (C) 2010 ARM Limited. All rights reserved. |
emilmont | 1:fdd22bb7aa52 | 3 | * |
emilmont | 1:fdd22bb7aa52 | 4 | * $Date: 15. February 2012 |
emilmont | 1:fdd22bb7aa52 | 5 | * $Revision: V1.1.0 |
emilmont | 1:fdd22bb7aa52 | 6 | * |
emilmont | 1:fdd22bb7aa52 | 7 | * Project: CMSIS DSP Library |
emilmont | 1:fdd22bb7aa52 | 8 | * Title: arm_correlate_q31.c |
emilmont | 1:fdd22bb7aa52 | 9 | * |
emilmont | 1:fdd22bb7aa52 | 10 | * Description: Correlation of Q31 sequences. |
emilmont | 1:fdd22bb7aa52 | 11 | * |
emilmont | 1:fdd22bb7aa52 | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
emilmont | 1:fdd22bb7aa52 | 13 | * |
emilmont | 1:fdd22bb7aa52 | 14 | * Version 1.1.0 2012/02/15 |
emilmont | 1:fdd22bb7aa52 | 15 | * Updated with more optimizations, bug fixes and minor API changes. |
emilmont | 1:fdd22bb7aa52 | 16 | * |
emilmont | 1:fdd22bb7aa52 | 17 | * Version 1.0.11 2011/10/18 |
emilmont | 1:fdd22bb7aa52 | 18 | * Bug Fix in conv, correlation, partial convolution. |
emilmont | 1:fdd22bb7aa52 | 19 | * |
emilmont | 1:fdd22bb7aa52 | 20 | * Version 1.0.10 2011/7/15 |
emilmont | 1:fdd22bb7aa52 | 21 | * Big Endian support added and Merged M0 and M3/M4 Source code. |
emilmont | 1:fdd22bb7aa52 | 22 | * |
emilmont | 1:fdd22bb7aa52 | 23 | * Version 1.0.3 2010/11/29 |
emilmont | 1:fdd22bb7aa52 | 24 | * Re-organized the CMSIS folders and updated documentation. |
emilmont | 1:fdd22bb7aa52 | 25 | * |
emilmont | 1:fdd22bb7aa52 | 26 | * Version 1.0.2 2010/11/11 |
emilmont | 1:fdd22bb7aa52 | 27 | * Documentation updated. |
emilmont | 1:fdd22bb7aa52 | 28 | * |
emilmont | 1:fdd22bb7aa52 | 29 | * Version 1.0.1 2010/10/05 |
emilmont | 1:fdd22bb7aa52 | 30 | * Production release and review comments incorporated. |
emilmont | 1:fdd22bb7aa52 | 31 | * |
emilmont | 1:fdd22bb7aa52 | 32 | * Version 1.0.0 2010/09/20 |
emilmont | 1:fdd22bb7aa52 | 33 | * Production release and review comments incorporated |
emilmont | 1:fdd22bb7aa52 | 34 | * |
emilmont | 1:fdd22bb7aa52 | 35 | * Version 0.0.7 2010/06/10 |
emilmont | 1:fdd22bb7aa52 | 36 | * Misra-C changes done |
emilmont | 1:fdd22bb7aa52 | 37 | * |
emilmont | 1:fdd22bb7aa52 | 38 | * -------------------------------------------------------------------- */ |
emilmont | 1:fdd22bb7aa52 | 39 | |
emilmont | 1:fdd22bb7aa52 | 40 | #include "arm_math.h" |
emilmont | 1:fdd22bb7aa52 | 41 | |
emilmont | 1:fdd22bb7aa52 | 42 | /** |
emilmont | 1:fdd22bb7aa52 | 43 | * @ingroup groupFilters |
emilmont | 1:fdd22bb7aa52 | 44 | */ |
emilmont | 1:fdd22bb7aa52 | 45 | |
emilmont | 1:fdd22bb7aa52 | 46 | /** |
emilmont | 1:fdd22bb7aa52 | 47 | * @addtogroup Corr |
emilmont | 1:fdd22bb7aa52 | 48 | * @{ |
emilmont | 1:fdd22bb7aa52 | 49 | */ |
emilmont | 1:fdd22bb7aa52 | 50 | |
emilmont | 1:fdd22bb7aa52 | 51 | /** |
emilmont | 1:fdd22bb7aa52 | 52 | * @brief Correlation of Q31 sequences. |
emilmont | 1:fdd22bb7aa52 | 53 | * @param[in] *pSrcA points to the first input sequence. |
emilmont | 1:fdd22bb7aa52 | 54 | * @param[in] srcALen length of the first input sequence. |
emilmont | 1:fdd22bb7aa52 | 55 | * @param[in] *pSrcB points to the second input sequence. |
emilmont | 1:fdd22bb7aa52 | 56 | * @param[in] srcBLen length of the second input sequence. |
emilmont | 1:fdd22bb7aa52 | 57 | * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1. |
emilmont | 1:fdd22bb7aa52 | 58 | * @return none. |
emilmont | 1:fdd22bb7aa52 | 59 | * |
emilmont | 1:fdd22bb7aa52 | 60 | * @details |
emilmont | 1:fdd22bb7aa52 | 61 | * <b>Scaling and Overflow Behavior:</b> |
emilmont | 1:fdd22bb7aa52 | 62 | * |
emilmont | 1:fdd22bb7aa52 | 63 | * \par |
emilmont | 1:fdd22bb7aa52 | 64 | * The function is implemented using an internal 64-bit accumulator. |
emilmont | 1:fdd22bb7aa52 | 65 | * The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit. |
emilmont | 1:fdd22bb7aa52 | 66 | * There is no saturation on intermediate additions. |
emilmont | 1:fdd22bb7aa52 | 67 | * Thus, if the accumulator overflows it wraps around and distorts the result. |
emilmont | 1:fdd22bb7aa52 | 68 | * The input signals should be scaled down to avoid intermediate overflows. |
emilmont | 1:fdd22bb7aa52 | 69 | * Scale down one of the inputs by 1/min(srcALen, srcBLen)to avoid overflows since a |
emilmont | 1:fdd22bb7aa52 | 70 | * maximum of min(srcALen, srcBLen) number of additions is carried internally. |
emilmont | 1:fdd22bb7aa52 | 71 | * The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result. |
emilmont | 1:fdd22bb7aa52 | 72 | * |
emilmont | 1:fdd22bb7aa52 | 73 | * \par |
emilmont | 1:fdd22bb7aa52 | 74 | * See <code>arm_correlate_fast_q31()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4. |
emilmont | 1:fdd22bb7aa52 | 75 | */ |
emilmont | 1:fdd22bb7aa52 | 76 | |
emilmont | 1:fdd22bb7aa52 | 77 | void arm_correlate_q31( |
emilmont | 1:fdd22bb7aa52 | 78 | q31_t * pSrcA, |
emilmont | 1:fdd22bb7aa52 | 79 | uint32_t srcALen, |
emilmont | 1:fdd22bb7aa52 | 80 | q31_t * pSrcB, |
emilmont | 1:fdd22bb7aa52 | 81 | uint32_t srcBLen, |
emilmont | 1:fdd22bb7aa52 | 82 | q31_t * pDst) |
emilmont | 1:fdd22bb7aa52 | 83 | { |
emilmont | 1:fdd22bb7aa52 | 84 | |
emilmont | 1:fdd22bb7aa52 | 85 | #ifndef ARM_MATH_CM0 |
emilmont | 1:fdd22bb7aa52 | 86 | |
emilmont | 1:fdd22bb7aa52 | 87 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
emilmont | 1:fdd22bb7aa52 | 88 | |
emilmont | 1:fdd22bb7aa52 | 89 | q31_t *pIn1; /* inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 90 | q31_t *pIn2; /* inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 91 | q31_t *pOut = pDst; /* output pointer */ |
emilmont | 1:fdd22bb7aa52 | 92 | q31_t *px; /* Intermediate inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 93 | q31_t *py; /* Intermediate inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 94 | q31_t *pSrc1; /* Intermediate pointers */ |
emilmont | 1:fdd22bb7aa52 | 95 | q63_t sum, acc0, acc1, acc2; /* Accumulators */ |
emilmont | 1:fdd22bb7aa52 | 96 | q31_t x0, x1, x2, c0; /* temporary variables for holding input and coefficient values */ |
emilmont | 1:fdd22bb7aa52 | 97 | uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */ |
emilmont | 1:fdd22bb7aa52 | 98 | int32_t inc = 1; /* Destination address modifier */ |
emilmont | 1:fdd22bb7aa52 | 99 | |
emilmont | 1:fdd22bb7aa52 | 100 | |
emilmont | 1:fdd22bb7aa52 | 101 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emilmont | 1:fdd22bb7aa52 | 102 | /* srcB is always made to slide across srcA. */ |
emilmont | 1:fdd22bb7aa52 | 103 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 104 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
emilmont | 1:fdd22bb7aa52 | 105 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 106 | /* and the destination pointer modifier, inc is set to -1 */ |
emilmont | 1:fdd22bb7aa52 | 107 | /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */ |
emilmont | 1:fdd22bb7aa52 | 108 | /* But to improve the performance, |
emilmont | 1:fdd22bb7aa52 | 109 | * we include zeroes in the output instead of zero padding either of the the inputs*/ |
emilmont | 1:fdd22bb7aa52 | 110 | /* If srcALen > srcBLen, |
emilmont | 1:fdd22bb7aa52 | 111 | * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 112 | /* If srcALen < srcBLen, |
emilmont | 1:fdd22bb7aa52 | 113 | * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 114 | if(srcALen >= srcBLen) |
emilmont | 1:fdd22bb7aa52 | 115 | { |
emilmont | 1:fdd22bb7aa52 | 116 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 117 | pIn1 = (pSrcA); |
emilmont | 1:fdd22bb7aa52 | 118 | |
emilmont | 1:fdd22bb7aa52 | 119 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 120 | pIn2 = (pSrcB); |
emilmont | 1:fdd22bb7aa52 | 121 | |
emilmont | 1:fdd22bb7aa52 | 122 | /* Number of output samples is calculated */ |
emilmont | 1:fdd22bb7aa52 | 123 | outBlockSize = (2u * srcALen) - 1u; |
emilmont | 1:fdd22bb7aa52 | 124 | |
emilmont | 1:fdd22bb7aa52 | 125 | /* When srcALen > srcBLen, zero padding is done to srcB |
emilmont | 1:fdd22bb7aa52 | 126 | * to make their lengths equal. |
emilmont | 1:fdd22bb7aa52 | 127 | * Instead, (outBlockSize - (srcALen + srcBLen - 1)) |
emilmont | 1:fdd22bb7aa52 | 128 | * number of output samples are made zero */ |
emilmont | 1:fdd22bb7aa52 | 129 | j = outBlockSize - (srcALen + (srcBLen - 1u)); |
emilmont | 1:fdd22bb7aa52 | 130 | |
emilmont | 1:fdd22bb7aa52 | 131 | /* Updating the pointer position to non zero value */ |
emilmont | 1:fdd22bb7aa52 | 132 | pOut += j; |
emilmont | 1:fdd22bb7aa52 | 133 | |
emilmont | 1:fdd22bb7aa52 | 134 | } |
emilmont | 1:fdd22bb7aa52 | 135 | else |
emilmont | 1:fdd22bb7aa52 | 136 | { |
emilmont | 1:fdd22bb7aa52 | 137 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 138 | pIn1 = (pSrcB); |
emilmont | 1:fdd22bb7aa52 | 139 | |
emilmont | 1:fdd22bb7aa52 | 140 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 141 | pIn2 = (pSrcA); |
emilmont | 1:fdd22bb7aa52 | 142 | |
emilmont | 1:fdd22bb7aa52 | 143 | /* srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 144 | j = srcBLen; |
emilmont | 1:fdd22bb7aa52 | 145 | srcBLen = srcALen; |
emilmont | 1:fdd22bb7aa52 | 146 | srcALen = j; |
emilmont | 1:fdd22bb7aa52 | 147 | |
emilmont | 1:fdd22bb7aa52 | 148 | /* CORR(x, y) = Reverse order(CORR(y, x)) */ |
emilmont | 1:fdd22bb7aa52 | 149 | /* Hence set the destination pointer to point to the last output sample */ |
emilmont | 1:fdd22bb7aa52 | 150 | pOut = pDst + ((srcALen + srcBLen) - 2u); |
emilmont | 1:fdd22bb7aa52 | 151 | |
emilmont | 1:fdd22bb7aa52 | 152 | /* Destination address modifier is set to -1 */ |
emilmont | 1:fdd22bb7aa52 | 153 | inc = -1; |
emilmont | 1:fdd22bb7aa52 | 154 | |
emilmont | 1:fdd22bb7aa52 | 155 | } |
emilmont | 1:fdd22bb7aa52 | 156 | |
emilmont | 1:fdd22bb7aa52 | 157 | /* The function is internally |
emilmont | 1:fdd22bb7aa52 | 158 | * divided into three parts according to the number of multiplications that has to be |
emilmont | 1:fdd22bb7aa52 | 159 | * taken place between inputA samples and inputB samples. In the first part of the |
emilmont | 1:fdd22bb7aa52 | 160 | * algorithm, the multiplications increase by one for every iteration. |
emilmont | 1:fdd22bb7aa52 | 161 | * In the second part of the algorithm, srcBLen number of multiplications are done. |
emilmont | 1:fdd22bb7aa52 | 162 | * In the third part of the algorithm, the multiplications decrease by one |
emilmont | 1:fdd22bb7aa52 | 163 | * for every iteration.*/ |
emilmont | 1:fdd22bb7aa52 | 164 | /* The algorithm is implemented in three stages. |
emilmont | 1:fdd22bb7aa52 | 165 | * The loop counters of each stage is initiated here. */ |
emilmont | 1:fdd22bb7aa52 | 166 | blockSize1 = srcBLen - 1u; |
emilmont | 1:fdd22bb7aa52 | 167 | blockSize2 = srcALen - (srcBLen - 1u); |
emilmont | 1:fdd22bb7aa52 | 168 | blockSize3 = blockSize1; |
emilmont | 1:fdd22bb7aa52 | 169 | |
emilmont | 1:fdd22bb7aa52 | 170 | /* -------------------------- |
emilmont | 1:fdd22bb7aa52 | 171 | * Initializations of stage1 |
emilmont | 1:fdd22bb7aa52 | 172 | * -------------------------*/ |
emilmont | 1:fdd22bb7aa52 | 173 | |
emilmont | 1:fdd22bb7aa52 | 174 | /* sum = x[0] * y[srcBlen - 1] |
emilmont | 1:fdd22bb7aa52 | 175 | * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1] |
emilmont | 1:fdd22bb7aa52 | 176 | * .... |
emilmont | 1:fdd22bb7aa52 | 177 | * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] |
emilmont | 1:fdd22bb7aa52 | 178 | */ |
emilmont | 1:fdd22bb7aa52 | 179 | |
emilmont | 1:fdd22bb7aa52 | 180 | /* In this stage the MAC operations are increased by 1 for every iteration. |
emilmont | 1:fdd22bb7aa52 | 181 | The count variable holds the number of MAC operations performed */ |
emilmont | 1:fdd22bb7aa52 | 182 | count = 1u; |
emilmont | 1:fdd22bb7aa52 | 183 | |
emilmont | 1:fdd22bb7aa52 | 184 | /* Working pointer of inputA */ |
emilmont | 1:fdd22bb7aa52 | 185 | px = pIn1; |
emilmont | 1:fdd22bb7aa52 | 186 | |
emilmont | 1:fdd22bb7aa52 | 187 | /* Working pointer of inputB */ |
emilmont | 1:fdd22bb7aa52 | 188 | pSrc1 = pIn2 + (srcBLen - 1u); |
emilmont | 1:fdd22bb7aa52 | 189 | py = pSrc1; |
emilmont | 1:fdd22bb7aa52 | 190 | |
emilmont | 1:fdd22bb7aa52 | 191 | /* ------------------------ |
emilmont | 1:fdd22bb7aa52 | 192 | * Stage1 process |
emilmont | 1:fdd22bb7aa52 | 193 | * ----------------------*/ |
emilmont | 1:fdd22bb7aa52 | 194 | |
emilmont | 1:fdd22bb7aa52 | 195 | /* The first stage starts here */ |
emilmont | 1:fdd22bb7aa52 | 196 | while(blockSize1 > 0u) |
emilmont | 1:fdd22bb7aa52 | 197 | { |
emilmont | 1:fdd22bb7aa52 | 198 | /* Accumulator is made zero for every iteration */ |
emilmont | 1:fdd22bb7aa52 | 199 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 200 | |
emilmont | 1:fdd22bb7aa52 | 201 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 202 | k = count >> 2; |
emilmont | 1:fdd22bb7aa52 | 203 | |
emilmont | 1:fdd22bb7aa52 | 204 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emilmont | 1:fdd22bb7aa52 | 205 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 206 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 207 | { |
emilmont | 1:fdd22bb7aa52 | 208 | /* x[0] * y[srcBLen - 4] */ |
emilmont | 1:fdd22bb7aa52 | 209 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 210 | /* x[1] * y[srcBLen - 3] */ |
emilmont | 1:fdd22bb7aa52 | 211 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 212 | /* x[2] * y[srcBLen - 2] */ |
emilmont | 1:fdd22bb7aa52 | 213 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 214 | /* x[3] * y[srcBLen - 1] */ |
emilmont | 1:fdd22bb7aa52 | 215 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 216 | |
emilmont | 1:fdd22bb7aa52 | 217 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 218 | k--; |
emilmont | 1:fdd22bb7aa52 | 219 | } |
emilmont | 1:fdd22bb7aa52 | 220 | |
emilmont | 1:fdd22bb7aa52 | 221 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
emilmont | 1:fdd22bb7aa52 | 222 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 223 | k = count % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 224 | |
emilmont | 1:fdd22bb7aa52 | 225 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 226 | { |
emilmont | 1:fdd22bb7aa52 | 227 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 228 | /* x[0] * y[srcBLen - 1] */ |
emilmont | 1:fdd22bb7aa52 | 229 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 230 | |
emilmont | 1:fdd22bb7aa52 | 231 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 232 | k--; |
emilmont | 1:fdd22bb7aa52 | 233 | } |
emilmont | 1:fdd22bb7aa52 | 234 | |
emilmont | 1:fdd22bb7aa52 | 235 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 236 | *pOut = (q31_t) (sum >> 31); |
emilmont | 1:fdd22bb7aa52 | 237 | /* Destination pointer is updated according to the address modifier, inc */ |
emilmont | 1:fdd22bb7aa52 | 238 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 239 | |
emilmont | 1:fdd22bb7aa52 | 240 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emilmont | 1:fdd22bb7aa52 | 241 | py = pSrc1 - count; |
emilmont | 1:fdd22bb7aa52 | 242 | px = pIn1; |
emilmont | 1:fdd22bb7aa52 | 243 | |
emilmont | 1:fdd22bb7aa52 | 244 | /* Increment the MAC count */ |
emilmont | 1:fdd22bb7aa52 | 245 | count++; |
emilmont | 1:fdd22bb7aa52 | 246 | |
emilmont | 1:fdd22bb7aa52 | 247 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 248 | blockSize1--; |
emilmont | 1:fdd22bb7aa52 | 249 | } |
emilmont | 1:fdd22bb7aa52 | 250 | |
emilmont | 1:fdd22bb7aa52 | 251 | /* -------------------------- |
emilmont | 1:fdd22bb7aa52 | 252 | * Initializations of stage2 |
emilmont | 1:fdd22bb7aa52 | 253 | * ------------------------*/ |
emilmont | 1:fdd22bb7aa52 | 254 | |
emilmont | 1:fdd22bb7aa52 | 255 | /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] |
emilmont | 1:fdd22bb7aa52 | 256 | * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] |
emilmont | 1:fdd22bb7aa52 | 257 | * .... |
emilmont | 1:fdd22bb7aa52 | 258 | * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emilmont | 1:fdd22bb7aa52 | 259 | */ |
emilmont | 1:fdd22bb7aa52 | 260 | |
emilmont | 1:fdd22bb7aa52 | 261 | /* Working pointer of inputA */ |
emilmont | 1:fdd22bb7aa52 | 262 | px = pIn1; |
emilmont | 1:fdd22bb7aa52 | 263 | |
emilmont | 1:fdd22bb7aa52 | 264 | /* Working pointer of inputB */ |
emilmont | 1:fdd22bb7aa52 | 265 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 266 | |
emilmont | 1:fdd22bb7aa52 | 267 | /* count is index by which the pointer pIn1 to be incremented */ |
emilmont | 1:fdd22bb7aa52 | 268 | count = 0u; |
emilmont | 1:fdd22bb7aa52 | 269 | |
emilmont | 1:fdd22bb7aa52 | 270 | /* ------------------- |
emilmont | 1:fdd22bb7aa52 | 271 | * Stage2 process |
emilmont | 1:fdd22bb7aa52 | 272 | * ------------------*/ |
emilmont | 1:fdd22bb7aa52 | 273 | |
emilmont | 1:fdd22bb7aa52 | 274 | /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. |
emilmont | 1:fdd22bb7aa52 | 275 | * So, to loop unroll over blockSize2, |
emilmont | 1:fdd22bb7aa52 | 276 | * srcBLen should be greater than or equal to 4 */ |
emilmont | 1:fdd22bb7aa52 | 277 | if(srcBLen >= 4u) |
emilmont | 1:fdd22bb7aa52 | 278 | { |
emilmont | 1:fdd22bb7aa52 | 279 | /* Loop unroll by 3 */ |
emilmont | 1:fdd22bb7aa52 | 280 | blkCnt = blockSize2 / 3; |
emilmont | 1:fdd22bb7aa52 | 281 | |
emilmont | 1:fdd22bb7aa52 | 282 | while(blkCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 283 | { |
emilmont | 1:fdd22bb7aa52 | 284 | /* Set all accumulators to zero */ |
emilmont | 1:fdd22bb7aa52 | 285 | acc0 = 0; |
emilmont | 1:fdd22bb7aa52 | 286 | acc1 = 0; |
emilmont | 1:fdd22bb7aa52 | 287 | acc2 = 0; |
emilmont | 1:fdd22bb7aa52 | 288 | |
emilmont | 1:fdd22bb7aa52 | 289 | /* read x[0], x[1] samples */ |
emilmont | 1:fdd22bb7aa52 | 290 | x0 = *(px++); |
emilmont | 1:fdd22bb7aa52 | 291 | x1 = *(px++); |
emilmont | 1:fdd22bb7aa52 | 292 | |
emilmont | 1:fdd22bb7aa52 | 293 | /* Apply loop unrolling and compute 3 MACs simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 294 | k = srcBLen / 3; |
emilmont | 1:fdd22bb7aa52 | 295 | |
emilmont | 1:fdd22bb7aa52 | 296 | /* First part of the processing with loop unrolling. Compute 3 MACs at a time. |
emilmont | 1:fdd22bb7aa52 | 297 | ** a second loop below computes MACs for the remaining 1 to 2 samples. */ |
emilmont | 1:fdd22bb7aa52 | 298 | do |
emilmont | 1:fdd22bb7aa52 | 299 | { |
emilmont | 1:fdd22bb7aa52 | 300 | /* Read y[0] sample */ |
emilmont | 1:fdd22bb7aa52 | 301 | c0 = *(py); |
emilmont | 1:fdd22bb7aa52 | 302 | |
emilmont | 1:fdd22bb7aa52 | 303 | /* Read x[2] sample */ |
emilmont | 1:fdd22bb7aa52 | 304 | x2 = *(px); |
emilmont | 1:fdd22bb7aa52 | 305 | |
emilmont | 1:fdd22bb7aa52 | 306 | /* Perform the multiply-accumulate */ |
emilmont | 1:fdd22bb7aa52 | 307 | /* acc0 += x[0] * y[0] */ |
emilmont | 1:fdd22bb7aa52 | 308 | acc0 += ((q63_t) x0 * c0); |
emilmont | 1:fdd22bb7aa52 | 309 | /* acc1 += x[1] * y[0] */ |
emilmont | 1:fdd22bb7aa52 | 310 | acc1 += ((q63_t) x1 * c0); |
emilmont | 1:fdd22bb7aa52 | 311 | /* acc2 += x[2] * y[0] */ |
emilmont | 1:fdd22bb7aa52 | 312 | acc2 += ((q63_t) x2 * c0); |
emilmont | 1:fdd22bb7aa52 | 313 | |
emilmont | 1:fdd22bb7aa52 | 314 | /* Read y[1] sample */ |
emilmont | 1:fdd22bb7aa52 | 315 | c0 = *(py + 1u); |
emilmont | 1:fdd22bb7aa52 | 316 | |
emilmont | 1:fdd22bb7aa52 | 317 | /* Read x[3] sample */ |
emilmont | 1:fdd22bb7aa52 | 318 | x0 = *(px + 1u); |
emilmont | 1:fdd22bb7aa52 | 319 | |
emilmont | 1:fdd22bb7aa52 | 320 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 321 | /* acc0 += x[1] * y[1] */ |
emilmont | 1:fdd22bb7aa52 | 322 | acc0 += ((q63_t) x1 * c0); |
emilmont | 1:fdd22bb7aa52 | 323 | /* acc1 += x[2] * y[1] */ |
emilmont | 1:fdd22bb7aa52 | 324 | acc1 += ((q63_t) x2 * c0); |
emilmont | 1:fdd22bb7aa52 | 325 | /* acc2 += x[3] * y[1] */ |
emilmont | 1:fdd22bb7aa52 | 326 | acc2 += ((q63_t) x0 * c0); |
emilmont | 1:fdd22bb7aa52 | 327 | |
emilmont | 1:fdd22bb7aa52 | 328 | /* Read y[2] sample */ |
emilmont | 1:fdd22bb7aa52 | 329 | c0 = *(py + 2u); |
emilmont | 1:fdd22bb7aa52 | 330 | |
emilmont | 1:fdd22bb7aa52 | 331 | /* Read x[4] sample */ |
emilmont | 1:fdd22bb7aa52 | 332 | x1 = *(px + 2u); |
emilmont | 1:fdd22bb7aa52 | 333 | |
emilmont | 1:fdd22bb7aa52 | 334 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 335 | /* acc0 += x[2] * y[2] */ |
emilmont | 1:fdd22bb7aa52 | 336 | acc0 += ((q63_t) x2 * c0); |
emilmont | 1:fdd22bb7aa52 | 337 | /* acc1 += x[3] * y[2] */ |
emilmont | 1:fdd22bb7aa52 | 338 | acc1 += ((q63_t) x0 * c0); |
emilmont | 1:fdd22bb7aa52 | 339 | /* acc2 += x[4] * y[2] */ |
emilmont | 1:fdd22bb7aa52 | 340 | acc2 += ((q63_t) x1 * c0); |
emilmont | 1:fdd22bb7aa52 | 341 | |
emilmont | 1:fdd22bb7aa52 | 342 | /* update scratch pointers */ |
emilmont | 1:fdd22bb7aa52 | 343 | px += 3u; |
emilmont | 1:fdd22bb7aa52 | 344 | py += 3u; |
emilmont | 1:fdd22bb7aa52 | 345 | |
emilmont | 1:fdd22bb7aa52 | 346 | } while(--k); |
emilmont | 1:fdd22bb7aa52 | 347 | |
emilmont | 1:fdd22bb7aa52 | 348 | /* If the srcBLen is not a multiple of 3, compute any remaining MACs here. |
emilmont | 1:fdd22bb7aa52 | 349 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 350 | k = srcBLen - (3 * (srcBLen / 3)); |
emilmont | 1:fdd22bb7aa52 | 351 | |
emilmont | 1:fdd22bb7aa52 | 352 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 353 | { |
emilmont | 1:fdd22bb7aa52 | 354 | /* Read y[4] sample */ |
emilmont | 1:fdd22bb7aa52 | 355 | c0 = *(py++); |
emilmont | 1:fdd22bb7aa52 | 356 | |
emilmont | 1:fdd22bb7aa52 | 357 | /* Read x[7] sample */ |
emilmont | 1:fdd22bb7aa52 | 358 | x2 = *(px++); |
emilmont | 1:fdd22bb7aa52 | 359 | |
emilmont | 1:fdd22bb7aa52 | 360 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 361 | /* acc0 += x[4] * y[4] */ |
emilmont | 1:fdd22bb7aa52 | 362 | acc0 += ((q63_t) x0 * c0); |
emilmont | 1:fdd22bb7aa52 | 363 | /* acc1 += x[5] * y[4] */ |
emilmont | 1:fdd22bb7aa52 | 364 | acc1 += ((q63_t) x1 * c0); |
emilmont | 1:fdd22bb7aa52 | 365 | /* acc2 += x[6] * y[4] */ |
emilmont | 1:fdd22bb7aa52 | 366 | acc2 += ((q63_t) x2 * c0); |
emilmont | 1:fdd22bb7aa52 | 367 | |
emilmont | 1:fdd22bb7aa52 | 368 | /* Reuse the present samples for the next MAC */ |
emilmont | 1:fdd22bb7aa52 | 369 | x0 = x1; |
emilmont | 1:fdd22bb7aa52 | 370 | x1 = x2; |
emilmont | 1:fdd22bb7aa52 | 371 | |
emilmont | 1:fdd22bb7aa52 | 372 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 373 | k--; |
emilmont | 1:fdd22bb7aa52 | 374 | } |
emilmont | 1:fdd22bb7aa52 | 375 | |
emilmont | 1:fdd22bb7aa52 | 376 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 377 | *pOut = (q31_t) (acc0 >> 31); |
emilmont | 1:fdd22bb7aa52 | 378 | /* Destination pointer is updated according to the address modifier, inc */ |
emilmont | 1:fdd22bb7aa52 | 379 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 380 | |
emilmont | 1:fdd22bb7aa52 | 381 | *pOut = (q31_t) (acc1 >> 31); |
emilmont | 1:fdd22bb7aa52 | 382 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 383 | |
emilmont | 1:fdd22bb7aa52 | 384 | *pOut = (q31_t) (acc2 >> 31); |
emilmont | 1:fdd22bb7aa52 | 385 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 386 | |
emilmont | 1:fdd22bb7aa52 | 387 | /* Increment the pointer pIn1 index, count by 3 */ |
emilmont | 1:fdd22bb7aa52 | 388 | count += 3u; |
emilmont | 1:fdd22bb7aa52 | 389 | |
emilmont | 1:fdd22bb7aa52 | 390 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emilmont | 1:fdd22bb7aa52 | 391 | px = pIn1 + count; |
emilmont | 1:fdd22bb7aa52 | 392 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 393 | |
emilmont | 1:fdd22bb7aa52 | 394 | |
emilmont | 1:fdd22bb7aa52 | 395 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 396 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 397 | } |
emilmont | 1:fdd22bb7aa52 | 398 | |
emilmont | 1:fdd22bb7aa52 | 399 | /* If the blockSize2 is not a multiple of 3, compute any remaining output samples here. |
emilmont | 1:fdd22bb7aa52 | 400 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 401 | blkCnt = blockSize2 - 3 * (blockSize2 / 3); |
emilmont | 1:fdd22bb7aa52 | 402 | |
emilmont | 1:fdd22bb7aa52 | 403 | while(blkCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 404 | { |
emilmont | 1:fdd22bb7aa52 | 405 | /* Accumulator is made zero for every iteration */ |
emilmont | 1:fdd22bb7aa52 | 406 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 407 | |
emilmont | 1:fdd22bb7aa52 | 408 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 409 | k = srcBLen >> 2u; |
emilmont | 1:fdd22bb7aa52 | 410 | |
emilmont | 1:fdd22bb7aa52 | 411 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emilmont | 1:fdd22bb7aa52 | 412 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 413 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 414 | { |
emilmont | 1:fdd22bb7aa52 | 415 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 416 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 417 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 418 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 419 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 420 | |
emilmont | 1:fdd22bb7aa52 | 421 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 422 | k--; |
emilmont | 1:fdd22bb7aa52 | 423 | } |
emilmont | 1:fdd22bb7aa52 | 424 | |
emilmont | 1:fdd22bb7aa52 | 425 | /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. |
emilmont | 1:fdd22bb7aa52 | 426 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 427 | k = srcBLen % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 428 | |
emilmont | 1:fdd22bb7aa52 | 429 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 430 | { |
emilmont | 1:fdd22bb7aa52 | 431 | /* Perform the multiply-accumulate */ |
emilmont | 1:fdd22bb7aa52 | 432 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 433 | |
emilmont | 1:fdd22bb7aa52 | 434 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 435 | k--; |
emilmont | 1:fdd22bb7aa52 | 436 | } |
emilmont | 1:fdd22bb7aa52 | 437 | |
emilmont | 1:fdd22bb7aa52 | 438 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 439 | *pOut = (q31_t) (sum >> 31); |
emilmont | 1:fdd22bb7aa52 | 440 | /* Destination pointer is updated according to the address modifier, inc */ |
emilmont | 1:fdd22bb7aa52 | 441 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 442 | |
emilmont | 1:fdd22bb7aa52 | 443 | /* Increment the MAC count */ |
emilmont | 1:fdd22bb7aa52 | 444 | count++; |
emilmont | 1:fdd22bb7aa52 | 445 | |
emilmont | 1:fdd22bb7aa52 | 446 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emilmont | 1:fdd22bb7aa52 | 447 | px = pIn1 + count; |
emilmont | 1:fdd22bb7aa52 | 448 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 449 | |
emilmont | 1:fdd22bb7aa52 | 450 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 451 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 452 | } |
emilmont | 1:fdd22bb7aa52 | 453 | } |
emilmont | 1:fdd22bb7aa52 | 454 | else |
emilmont | 1:fdd22bb7aa52 | 455 | { |
emilmont | 1:fdd22bb7aa52 | 456 | /* If the srcBLen is not a multiple of 4, |
emilmont | 1:fdd22bb7aa52 | 457 | * the blockSize2 loop cannot be unrolled by 4 */ |
emilmont | 1:fdd22bb7aa52 | 458 | blkCnt = blockSize2; |
emilmont | 1:fdd22bb7aa52 | 459 | |
emilmont | 1:fdd22bb7aa52 | 460 | while(blkCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 461 | { |
emilmont | 1:fdd22bb7aa52 | 462 | /* Accumulator is made zero for every iteration */ |
emilmont | 1:fdd22bb7aa52 | 463 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 464 | |
emilmont | 1:fdd22bb7aa52 | 465 | /* Loop over srcBLen */ |
emilmont | 1:fdd22bb7aa52 | 466 | k = srcBLen; |
emilmont | 1:fdd22bb7aa52 | 467 | |
emilmont | 1:fdd22bb7aa52 | 468 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 469 | { |
emilmont | 1:fdd22bb7aa52 | 470 | /* Perform the multiply-accumulate */ |
emilmont | 1:fdd22bb7aa52 | 471 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 472 | |
emilmont | 1:fdd22bb7aa52 | 473 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 474 | k--; |
emilmont | 1:fdd22bb7aa52 | 475 | } |
emilmont | 1:fdd22bb7aa52 | 476 | |
emilmont | 1:fdd22bb7aa52 | 477 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 478 | *pOut = (q31_t) (sum >> 31); |
emilmont | 1:fdd22bb7aa52 | 479 | /* Destination pointer is updated according to the address modifier, inc */ |
emilmont | 1:fdd22bb7aa52 | 480 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 481 | |
emilmont | 1:fdd22bb7aa52 | 482 | /* Increment the MAC count */ |
emilmont | 1:fdd22bb7aa52 | 483 | count++; |
emilmont | 1:fdd22bb7aa52 | 484 | |
emilmont | 1:fdd22bb7aa52 | 485 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emilmont | 1:fdd22bb7aa52 | 486 | px = pIn1 + count; |
emilmont | 1:fdd22bb7aa52 | 487 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 488 | |
emilmont | 1:fdd22bb7aa52 | 489 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 490 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 491 | } |
emilmont | 1:fdd22bb7aa52 | 492 | } |
emilmont | 1:fdd22bb7aa52 | 493 | |
emilmont | 1:fdd22bb7aa52 | 494 | /* -------------------------- |
emilmont | 1:fdd22bb7aa52 | 495 | * Initializations of stage3 |
emilmont | 1:fdd22bb7aa52 | 496 | * -------------------------*/ |
emilmont | 1:fdd22bb7aa52 | 497 | |
emilmont | 1:fdd22bb7aa52 | 498 | /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emilmont | 1:fdd22bb7aa52 | 499 | * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emilmont | 1:fdd22bb7aa52 | 500 | * .... |
emilmont | 1:fdd22bb7aa52 | 501 | * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] |
emilmont | 1:fdd22bb7aa52 | 502 | * sum += x[srcALen-1] * y[0] |
emilmont | 1:fdd22bb7aa52 | 503 | */ |
emilmont | 1:fdd22bb7aa52 | 504 | |
emilmont | 1:fdd22bb7aa52 | 505 | /* In this stage the MAC operations are decreased by 1 for every iteration. |
emilmont | 1:fdd22bb7aa52 | 506 | The count variable holds the number of MAC operations performed */ |
emilmont | 1:fdd22bb7aa52 | 507 | count = srcBLen - 1u; |
emilmont | 1:fdd22bb7aa52 | 508 | |
emilmont | 1:fdd22bb7aa52 | 509 | /* Working pointer of inputA */ |
emilmont | 1:fdd22bb7aa52 | 510 | pSrc1 = pIn1 + (srcALen - (srcBLen - 1u)); |
emilmont | 1:fdd22bb7aa52 | 511 | px = pSrc1; |
emilmont | 1:fdd22bb7aa52 | 512 | |
emilmont | 1:fdd22bb7aa52 | 513 | /* Working pointer of inputB */ |
emilmont | 1:fdd22bb7aa52 | 514 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 515 | |
emilmont | 1:fdd22bb7aa52 | 516 | /* ------------------- |
emilmont | 1:fdd22bb7aa52 | 517 | * Stage3 process |
emilmont | 1:fdd22bb7aa52 | 518 | * ------------------*/ |
emilmont | 1:fdd22bb7aa52 | 519 | |
emilmont | 1:fdd22bb7aa52 | 520 | while(blockSize3 > 0u) |
emilmont | 1:fdd22bb7aa52 | 521 | { |
emilmont | 1:fdd22bb7aa52 | 522 | /* Accumulator is made zero for every iteration */ |
emilmont | 1:fdd22bb7aa52 | 523 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 524 | |
emilmont | 1:fdd22bb7aa52 | 525 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 526 | k = count >> 2u; |
emilmont | 1:fdd22bb7aa52 | 527 | |
emilmont | 1:fdd22bb7aa52 | 528 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emilmont | 1:fdd22bb7aa52 | 529 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 530 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 531 | { |
emilmont | 1:fdd22bb7aa52 | 532 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 533 | /* sum += x[srcALen - srcBLen + 4] * y[3] */ |
emilmont | 1:fdd22bb7aa52 | 534 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 535 | /* sum += x[srcALen - srcBLen + 3] * y[2] */ |
emilmont | 1:fdd22bb7aa52 | 536 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 537 | /* sum += x[srcALen - srcBLen + 2] * y[1] */ |
emilmont | 1:fdd22bb7aa52 | 538 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 539 | /* sum += x[srcALen - srcBLen + 1] * y[0] */ |
emilmont | 1:fdd22bb7aa52 | 540 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 541 | |
emilmont | 1:fdd22bb7aa52 | 542 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 543 | k--; |
emilmont | 1:fdd22bb7aa52 | 544 | } |
emilmont | 1:fdd22bb7aa52 | 545 | |
emilmont | 1:fdd22bb7aa52 | 546 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
emilmont | 1:fdd22bb7aa52 | 547 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 548 | k = count % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 549 | |
emilmont | 1:fdd22bb7aa52 | 550 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 551 | { |
emilmont | 1:fdd22bb7aa52 | 552 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 553 | sum += (q63_t) * px++ * (*py++); |
emilmont | 1:fdd22bb7aa52 | 554 | |
emilmont | 1:fdd22bb7aa52 | 555 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 556 | k--; |
emilmont | 1:fdd22bb7aa52 | 557 | } |
emilmont | 1:fdd22bb7aa52 | 558 | |
emilmont | 1:fdd22bb7aa52 | 559 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 560 | *pOut = (q31_t) (sum >> 31); |
emilmont | 1:fdd22bb7aa52 | 561 | /* Destination pointer is updated according to the address modifier, inc */ |
emilmont | 1:fdd22bb7aa52 | 562 | pOut += inc; |
emilmont | 1:fdd22bb7aa52 | 563 | |
emilmont | 1:fdd22bb7aa52 | 564 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emilmont | 1:fdd22bb7aa52 | 565 | px = ++pSrc1; |
emilmont | 1:fdd22bb7aa52 | 566 | py = pIn2; |
emilmont | 1:fdd22bb7aa52 | 567 | |
emilmont | 1:fdd22bb7aa52 | 568 | /* Decrement the MAC count */ |
emilmont | 1:fdd22bb7aa52 | 569 | count--; |
emilmont | 1:fdd22bb7aa52 | 570 | |
emilmont | 1:fdd22bb7aa52 | 571 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 572 | blockSize3--; |
emilmont | 1:fdd22bb7aa52 | 573 | } |
emilmont | 1:fdd22bb7aa52 | 574 | |
emilmont | 1:fdd22bb7aa52 | 575 | #else |
emilmont | 1:fdd22bb7aa52 | 576 | |
emilmont | 1:fdd22bb7aa52 | 577 | /* Run the below code for Cortex-M0 */ |
emilmont | 1:fdd22bb7aa52 | 578 | |
emilmont | 1:fdd22bb7aa52 | 579 | q31_t *pIn1 = pSrcA; /* inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 580 | q31_t *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 581 | q63_t sum; /* Accumulators */ |
emilmont | 1:fdd22bb7aa52 | 582 | uint32_t i = 0u, j; /* loop counters */ |
emilmont | 1:fdd22bb7aa52 | 583 | uint32_t inv = 0u; /* Reverse order flag */ |
emilmont | 1:fdd22bb7aa52 | 584 | uint32_t tot = 0u; /* Length */ |
emilmont | 1:fdd22bb7aa52 | 585 | |
emilmont | 1:fdd22bb7aa52 | 586 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emilmont | 1:fdd22bb7aa52 | 587 | /* srcB is always made to slide across srcA. */ |
emilmont | 1:fdd22bb7aa52 | 588 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 589 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
emilmont | 1:fdd22bb7aa52 | 590 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 591 | /* and a varaible, inv is set to 1 */ |
emilmont | 1:fdd22bb7aa52 | 592 | /* If lengths are not equal then zero pad has to be done to make the two |
emilmont | 1:fdd22bb7aa52 | 593 | * inputs of same length. But to improve the performance, we include zeroes |
emilmont | 1:fdd22bb7aa52 | 594 | * in the output instead of zero padding either of the the inputs*/ |
emilmont | 1:fdd22bb7aa52 | 595 | /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the |
emilmont | 1:fdd22bb7aa52 | 596 | * starting of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 597 | /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the |
emilmont | 1:fdd22bb7aa52 | 598 | * ending of the output buffer */ |
emilmont | 1:fdd22bb7aa52 | 599 | /* Once the zero padding is done the remaining of the output is calcualted |
emilmont | 1:fdd22bb7aa52 | 600 | * using correlation but with the shorter signal time shifted. */ |
emilmont | 1:fdd22bb7aa52 | 601 | |
emilmont | 1:fdd22bb7aa52 | 602 | /* Calculate the length of the remaining sequence */ |
emilmont | 1:fdd22bb7aa52 | 603 | tot = ((srcALen + srcBLen) - 2u); |
emilmont | 1:fdd22bb7aa52 | 604 | |
emilmont | 1:fdd22bb7aa52 | 605 | if(srcALen > srcBLen) |
emilmont | 1:fdd22bb7aa52 | 606 | { |
emilmont | 1:fdd22bb7aa52 | 607 | /* Calculating the number of zeros to be padded to the output */ |
emilmont | 1:fdd22bb7aa52 | 608 | j = srcALen - srcBLen; |
emilmont | 1:fdd22bb7aa52 | 609 | |
emilmont | 1:fdd22bb7aa52 | 610 | /* Initialise the pointer after zero padding */ |
emilmont | 1:fdd22bb7aa52 | 611 | pDst += j; |
emilmont | 1:fdd22bb7aa52 | 612 | } |
emilmont | 1:fdd22bb7aa52 | 613 | |
emilmont | 1:fdd22bb7aa52 | 614 | else if(srcALen < srcBLen) |
emilmont | 1:fdd22bb7aa52 | 615 | { |
emilmont | 1:fdd22bb7aa52 | 616 | /* Initialization to inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 617 | pIn1 = pSrcB; |
emilmont | 1:fdd22bb7aa52 | 618 | |
emilmont | 1:fdd22bb7aa52 | 619 | /* Initialization to the end of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 620 | pIn2 = pSrcA + (srcALen - 1u); |
emilmont | 1:fdd22bb7aa52 | 621 | |
emilmont | 1:fdd22bb7aa52 | 622 | /* Initialisation of the pointer after zero padding */ |
emilmont | 1:fdd22bb7aa52 | 623 | pDst = pDst + tot; |
emilmont | 1:fdd22bb7aa52 | 624 | |
emilmont | 1:fdd22bb7aa52 | 625 | /* Swapping the lengths */ |
emilmont | 1:fdd22bb7aa52 | 626 | j = srcALen; |
emilmont | 1:fdd22bb7aa52 | 627 | srcALen = srcBLen; |
emilmont | 1:fdd22bb7aa52 | 628 | srcBLen = j; |
emilmont | 1:fdd22bb7aa52 | 629 | |
emilmont | 1:fdd22bb7aa52 | 630 | /* Setting the reverse flag */ |
emilmont | 1:fdd22bb7aa52 | 631 | inv = 1; |
emilmont | 1:fdd22bb7aa52 | 632 | |
emilmont | 1:fdd22bb7aa52 | 633 | } |
emilmont | 1:fdd22bb7aa52 | 634 | |
emilmont | 1:fdd22bb7aa52 | 635 | /* Loop to calculate correlation for output length number of times */ |
emilmont | 1:fdd22bb7aa52 | 636 | for (i = 0u; i <= tot; i++) |
emilmont | 1:fdd22bb7aa52 | 637 | { |
emilmont | 1:fdd22bb7aa52 | 638 | /* Initialize sum with zero to carry on MAC operations */ |
emilmont | 1:fdd22bb7aa52 | 639 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 640 | |
emilmont | 1:fdd22bb7aa52 | 641 | /* Loop to perform MAC operations according to correlation equation */ |
emilmont | 1:fdd22bb7aa52 | 642 | for (j = 0u; j <= i; j++) |
emilmont | 1:fdd22bb7aa52 | 643 | { |
emilmont | 1:fdd22bb7aa52 | 644 | /* Check the array limitations */ |
emilmont | 1:fdd22bb7aa52 | 645 | if((((i - j) < srcBLen) && (j < srcALen))) |
emilmont | 1:fdd22bb7aa52 | 646 | { |
emilmont | 1:fdd22bb7aa52 | 647 | /* z[i] += x[i-j] * y[j] */ |
emilmont | 1:fdd22bb7aa52 | 648 | sum += ((q63_t) pIn1[j] * pIn2[-((int32_t) i - j)]); |
emilmont | 1:fdd22bb7aa52 | 649 | } |
emilmont | 1:fdd22bb7aa52 | 650 | } |
emilmont | 1:fdd22bb7aa52 | 651 | /* Store the output in the destination buffer */ |
emilmont | 1:fdd22bb7aa52 | 652 | if(inv == 1) |
emilmont | 1:fdd22bb7aa52 | 653 | *pDst-- = (q31_t) (sum >> 31u); |
emilmont | 1:fdd22bb7aa52 | 654 | else |
emilmont | 1:fdd22bb7aa52 | 655 | *pDst++ = (q31_t) (sum >> 31u); |
emilmont | 1:fdd22bb7aa52 | 656 | } |
emilmont | 1:fdd22bb7aa52 | 657 | |
emilmont | 1:fdd22bb7aa52 | 658 | #endif /* #ifndef ARM_MATH_CM0 */ |
emilmont | 1:fdd22bb7aa52 | 659 | |
emilmont | 1:fdd22bb7aa52 | 660 | } |
emilmont | 1:fdd22bb7aa52 | 661 | |
emilmont | 1:fdd22bb7aa52 | 662 | /** |
emilmont | 1:fdd22bb7aa52 | 663 | * @} end of Corr group |
emilmont | 1:fdd22bb7aa52 | 664 | */ |