V4.0.1 of the ARM CMSIS DSP libraries. Note that arm_bitreversal2.s, arm_cfft_f32.c and arm_rfft_fast_f32.c had to be removed. arm_bitreversal2.s will not assemble with the online tools. So, the fast f32 FFT functions are not yet available. All the other FFT functions are available.
Dependents: MPU9150_Example fir_f32 fir_f32 MPU9150_nucleo_noni2cdev ... more
FilteringFunctions/arm_correlate_q7.c@0:3d9c67d97d6f, 2014-07-28 (annotated)
- Committer:
- emh203
- Date:
- Mon Jul 28 15:03:15 2014 +0000
- Revision:
- 0:3d9c67d97d6f
1st working commit. Had to remove arm_bitreversal2.s arm_cfft_f32.c and arm_rfft_fast_f32.c. The .s will not assemble. For now I removed these functions so we could at least have a library for the other functions.
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emh203 | 0:3d9c67d97d6f | 1 | /* ---------------------------------------------------------------------- |
emh203 | 0:3d9c67d97d6f | 2 | * Copyright (C) 2010-2014 ARM Limited. All rights reserved. |
emh203 | 0:3d9c67d97d6f | 3 | * |
emh203 | 0:3d9c67d97d6f | 4 | * $Date: 12. March 2014 |
emh203 | 0:3d9c67d97d6f | 5 | * $Revision: V1.4.3 |
emh203 | 0:3d9c67d97d6f | 6 | * |
emh203 | 0:3d9c67d97d6f | 7 | * Project: CMSIS DSP Library |
emh203 | 0:3d9c67d97d6f | 8 | * Title: arm_correlate_q7.c |
emh203 | 0:3d9c67d97d6f | 9 | * |
emh203 | 0:3d9c67d97d6f | 10 | * Description: Correlation of Q7 sequences. |
emh203 | 0:3d9c67d97d6f | 11 | * |
emh203 | 0:3d9c67d97d6f | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
emh203 | 0:3d9c67d97d6f | 13 | * |
emh203 | 0:3d9c67d97d6f | 14 | * Redistribution and use in source and binary forms, with or without |
emh203 | 0:3d9c67d97d6f | 15 | * modification, are permitted provided that the following conditions |
emh203 | 0:3d9c67d97d6f | 16 | * are met: |
emh203 | 0:3d9c67d97d6f | 17 | * - Redistributions of source code must retain the above copyright |
emh203 | 0:3d9c67d97d6f | 18 | * notice, this list of conditions and the following disclaimer. |
emh203 | 0:3d9c67d97d6f | 19 | * - Redistributions in binary form must reproduce the above copyright |
emh203 | 0:3d9c67d97d6f | 20 | * notice, this list of conditions and the following disclaimer in |
emh203 | 0:3d9c67d97d6f | 21 | * the documentation and/or other materials provided with the |
emh203 | 0:3d9c67d97d6f | 22 | * distribution. |
emh203 | 0:3d9c67d97d6f | 23 | * - Neither the name of ARM LIMITED nor the names of its contributors |
emh203 | 0:3d9c67d97d6f | 24 | * may be used to endorse or promote products derived from this |
emh203 | 0:3d9c67d97d6f | 25 | * software without specific prior written permission. |
emh203 | 0:3d9c67d97d6f | 26 | * |
emh203 | 0:3d9c67d97d6f | 27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
emh203 | 0:3d9c67d97d6f | 28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
emh203 | 0:3d9c67d97d6f | 29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
emh203 | 0:3d9c67d97d6f | 30 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
emh203 | 0:3d9c67d97d6f | 31 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
emh203 | 0:3d9c67d97d6f | 32 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
emh203 | 0:3d9c67d97d6f | 33 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
emh203 | 0:3d9c67d97d6f | 34 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
emh203 | 0:3d9c67d97d6f | 35 | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
emh203 | 0:3d9c67d97d6f | 36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
emh203 | 0:3d9c67d97d6f | 37 | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
emh203 | 0:3d9c67d97d6f | 38 | * POSSIBILITY OF SUCH DAMAGE. |
emh203 | 0:3d9c67d97d6f | 39 | * -------------------------------------------------------------------- */ |
emh203 | 0:3d9c67d97d6f | 40 | |
emh203 | 0:3d9c67d97d6f | 41 | #include "arm_math.h" |
emh203 | 0:3d9c67d97d6f | 42 | |
emh203 | 0:3d9c67d97d6f | 43 | /** |
emh203 | 0:3d9c67d97d6f | 44 | * @ingroup groupFilters |
emh203 | 0:3d9c67d97d6f | 45 | */ |
emh203 | 0:3d9c67d97d6f | 46 | |
emh203 | 0:3d9c67d97d6f | 47 | /** |
emh203 | 0:3d9c67d97d6f | 48 | * @addtogroup Corr |
emh203 | 0:3d9c67d97d6f | 49 | * @{ |
emh203 | 0:3d9c67d97d6f | 50 | */ |
emh203 | 0:3d9c67d97d6f | 51 | |
emh203 | 0:3d9c67d97d6f | 52 | /** |
emh203 | 0:3d9c67d97d6f | 53 | * @brief Correlation of Q7 sequences. |
emh203 | 0:3d9c67d97d6f | 54 | * @param[in] *pSrcA points to the first input sequence. |
emh203 | 0:3d9c67d97d6f | 55 | * @param[in] srcALen length of the first input sequence. |
emh203 | 0:3d9c67d97d6f | 56 | * @param[in] *pSrcB points to the second input sequence. |
emh203 | 0:3d9c67d97d6f | 57 | * @param[in] srcBLen length of the second input sequence. |
emh203 | 0:3d9c67d97d6f | 58 | * @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1. |
emh203 | 0:3d9c67d97d6f | 59 | * @return none. |
emh203 | 0:3d9c67d97d6f | 60 | * |
emh203 | 0:3d9c67d97d6f | 61 | * @details |
emh203 | 0:3d9c67d97d6f | 62 | * <b>Scaling and Overflow Behavior:</b> |
emh203 | 0:3d9c67d97d6f | 63 | * |
emh203 | 0:3d9c67d97d6f | 64 | * \par |
emh203 | 0:3d9c67d97d6f | 65 | * The function is implemented using a 32-bit internal accumulator. |
emh203 | 0:3d9c67d97d6f | 66 | * Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result. |
emh203 | 0:3d9c67d97d6f | 67 | * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format. |
emh203 | 0:3d9c67d97d6f | 68 | * This approach provides 17 guard bits and there is no risk of overflow as long as <code>max(srcALen, srcBLen)<131072</code>. |
emh203 | 0:3d9c67d97d6f | 69 | * The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and saturated to 1.7 format. |
emh203 | 0:3d9c67d97d6f | 70 | * |
emh203 | 0:3d9c67d97d6f | 71 | * \par |
emh203 | 0:3d9c67d97d6f | 72 | * Refer the function <code>arm_correlate_opt_q7()</code> for a faster implementation of this function. |
emh203 | 0:3d9c67d97d6f | 73 | * |
emh203 | 0:3d9c67d97d6f | 74 | */ |
emh203 | 0:3d9c67d97d6f | 75 | |
emh203 | 0:3d9c67d97d6f | 76 | void arm_correlate_q7( |
emh203 | 0:3d9c67d97d6f | 77 | q7_t * pSrcA, |
emh203 | 0:3d9c67d97d6f | 78 | uint32_t srcALen, |
emh203 | 0:3d9c67d97d6f | 79 | q7_t * pSrcB, |
emh203 | 0:3d9c67d97d6f | 80 | uint32_t srcBLen, |
emh203 | 0:3d9c67d97d6f | 81 | q7_t * pDst) |
emh203 | 0:3d9c67d97d6f | 82 | { |
emh203 | 0:3d9c67d97d6f | 83 | |
emh203 | 0:3d9c67d97d6f | 84 | |
emh203 | 0:3d9c67d97d6f | 85 | #ifndef ARM_MATH_CM0_FAMILY |
emh203 | 0:3d9c67d97d6f | 86 | |
emh203 | 0:3d9c67d97d6f | 87 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
emh203 | 0:3d9c67d97d6f | 88 | |
emh203 | 0:3d9c67d97d6f | 89 | q7_t *pIn1; /* inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 90 | q7_t *pIn2; /* inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 91 | q7_t *pOut = pDst; /* output pointer */ |
emh203 | 0:3d9c67d97d6f | 92 | q7_t *px; /* Intermediate inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 93 | q7_t *py; /* Intermediate inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 94 | q7_t *pSrc1; /* Intermediate pointers */ |
emh203 | 0:3d9c67d97d6f | 95 | q31_t sum, acc0, acc1, acc2, acc3; /* Accumulators */ |
emh203 | 0:3d9c67d97d6f | 96 | q31_t input1, input2; /* temporary variables */ |
emh203 | 0:3d9c67d97d6f | 97 | q15_t in1, in2; /* temporary variables */ |
emh203 | 0:3d9c67d97d6f | 98 | q7_t x0, x1, x2, x3, c0, c1; /* temporary variables for holding input and coefficient values */ |
emh203 | 0:3d9c67d97d6f | 99 | uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counter */ |
emh203 | 0:3d9c67d97d6f | 100 | int32_t inc = 1; |
emh203 | 0:3d9c67d97d6f | 101 | |
emh203 | 0:3d9c67d97d6f | 102 | |
emh203 | 0:3d9c67d97d6f | 103 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emh203 | 0:3d9c67d97d6f | 104 | /* srcB is always made to slide across srcA. */ |
emh203 | 0:3d9c67d97d6f | 105 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emh203 | 0:3d9c67d97d6f | 106 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
emh203 | 0:3d9c67d97d6f | 107 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 108 | /* and the destination pointer modifier, inc is set to -1 */ |
emh203 | 0:3d9c67d97d6f | 109 | /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */ |
emh203 | 0:3d9c67d97d6f | 110 | /* But to improve the performance, |
emh203 | 0:3d9c67d97d6f | 111 | * we include zeroes in the output instead of zero padding either of the the inputs*/ |
emh203 | 0:3d9c67d97d6f | 112 | /* If srcALen > srcBLen, |
emh203 | 0:3d9c67d97d6f | 113 | * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 114 | /* If srcALen < srcBLen, |
emh203 | 0:3d9c67d97d6f | 115 | * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 116 | if(srcALen >= srcBLen) |
emh203 | 0:3d9c67d97d6f | 117 | { |
emh203 | 0:3d9c67d97d6f | 118 | /* Initialization of inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 119 | pIn1 = (pSrcA); |
emh203 | 0:3d9c67d97d6f | 120 | |
emh203 | 0:3d9c67d97d6f | 121 | /* Initialization of inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 122 | pIn2 = (pSrcB); |
emh203 | 0:3d9c67d97d6f | 123 | |
emh203 | 0:3d9c67d97d6f | 124 | /* Number of output samples is calculated */ |
emh203 | 0:3d9c67d97d6f | 125 | outBlockSize = (2u * srcALen) - 1u; |
emh203 | 0:3d9c67d97d6f | 126 | |
emh203 | 0:3d9c67d97d6f | 127 | /* When srcALen > srcBLen, zero padding is done to srcB |
emh203 | 0:3d9c67d97d6f | 128 | * to make their lengths equal. |
emh203 | 0:3d9c67d97d6f | 129 | * Instead, (outBlockSize - (srcALen + srcBLen - 1)) |
emh203 | 0:3d9c67d97d6f | 130 | * number of output samples are made zero */ |
emh203 | 0:3d9c67d97d6f | 131 | j = outBlockSize - (srcALen + (srcBLen - 1u)); |
emh203 | 0:3d9c67d97d6f | 132 | |
emh203 | 0:3d9c67d97d6f | 133 | /* Updating the pointer position to non zero value */ |
emh203 | 0:3d9c67d97d6f | 134 | pOut += j; |
emh203 | 0:3d9c67d97d6f | 135 | |
emh203 | 0:3d9c67d97d6f | 136 | } |
emh203 | 0:3d9c67d97d6f | 137 | else |
emh203 | 0:3d9c67d97d6f | 138 | { |
emh203 | 0:3d9c67d97d6f | 139 | /* Initialization of inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 140 | pIn1 = (pSrcB); |
emh203 | 0:3d9c67d97d6f | 141 | |
emh203 | 0:3d9c67d97d6f | 142 | /* Initialization of inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 143 | pIn2 = (pSrcA); |
emh203 | 0:3d9c67d97d6f | 144 | |
emh203 | 0:3d9c67d97d6f | 145 | /* srcBLen is always considered as shorter or equal to srcALen */ |
emh203 | 0:3d9c67d97d6f | 146 | j = srcBLen; |
emh203 | 0:3d9c67d97d6f | 147 | srcBLen = srcALen; |
emh203 | 0:3d9c67d97d6f | 148 | srcALen = j; |
emh203 | 0:3d9c67d97d6f | 149 | |
emh203 | 0:3d9c67d97d6f | 150 | /* CORR(x, y) = Reverse order(CORR(y, x)) */ |
emh203 | 0:3d9c67d97d6f | 151 | /* Hence set the destination pointer to point to the last output sample */ |
emh203 | 0:3d9c67d97d6f | 152 | pOut = pDst + ((srcALen + srcBLen) - 2u); |
emh203 | 0:3d9c67d97d6f | 153 | |
emh203 | 0:3d9c67d97d6f | 154 | /* Destination address modifier is set to -1 */ |
emh203 | 0:3d9c67d97d6f | 155 | inc = -1; |
emh203 | 0:3d9c67d97d6f | 156 | |
emh203 | 0:3d9c67d97d6f | 157 | } |
emh203 | 0:3d9c67d97d6f | 158 | |
emh203 | 0:3d9c67d97d6f | 159 | /* The function is internally |
emh203 | 0:3d9c67d97d6f | 160 | * divided into three parts according to the number of multiplications that has to be |
emh203 | 0:3d9c67d97d6f | 161 | * taken place between inputA samples and inputB samples. In the first part of the |
emh203 | 0:3d9c67d97d6f | 162 | * algorithm, the multiplications increase by one for every iteration. |
emh203 | 0:3d9c67d97d6f | 163 | * In the second part of the algorithm, srcBLen number of multiplications are done. |
emh203 | 0:3d9c67d97d6f | 164 | * In the third part of the algorithm, the multiplications decrease by one |
emh203 | 0:3d9c67d97d6f | 165 | * for every iteration.*/ |
emh203 | 0:3d9c67d97d6f | 166 | /* The algorithm is implemented in three stages. |
emh203 | 0:3d9c67d97d6f | 167 | * The loop counters of each stage is initiated here. */ |
emh203 | 0:3d9c67d97d6f | 168 | blockSize1 = srcBLen - 1u; |
emh203 | 0:3d9c67d97d6f | 169 | blockSize2 = srcALen - (srcBLen - 1u); |
emh203 | 0:3d9c67d97d6f | 170 | blockSize3 = blockSize1; |
emh203 | 0:3d9c67d97d6f | 171 | |
emh203 | 0:3d9c67d97d6f | 172 | /* -------------------------- |
emh203 | 0:3d9c67d97d6f | 173 | * Initializations of stage1 |
emh203 | 0:3d9c67d97d6f | 174 | * -------------------------*/ |
emh203 | 0:3d9c67d97d6f | 175 | |
emh203 | 0:3d9c67d97d6f | 176 | /* sum = x[0] * y[srcBlen - 1] |
emh203 | 0:3d9c67d97d6f | 177 | * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1] |
emh203 | 0:3d9c67d97d6f | 178 | * .... |
emh203 | 0:3d9c67d97d6f | 179 | * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] |
emh203 | 0:3d9c67d97d6f | 180 | */ |
emh203 | 0:3d9c67d97d6f | 181 | |
emh203 | 0:3d9c67d97d6f | 182 | /* In this stage the MAC operations are increased by 1 for every iteration. |
emh203 | 0:3d9c67d97d6f | 183 | The count variable holds the number of MAC operations performed */ |
emh203 | 0:3d9c67d97d6f | 184 | count = 1u; |
emh203 | 0:3d9c67d97d6f | 185 | |
emh203 | 0:3d9c67d97d6f | 186 | /* Working pointer of inputA */ |
emh203 | 0:3d9c67d97d6f | 187 | px = pIn1; |
emh203 | 0:3d9c67d97d6f | 188 | |
emh203 | 0:3d9c67d97d6f | 189 | /* Working pointer of inputB */ |
emh203 | 0:3d9c67d97d6f | 190 | pSrc1 = pIn2 + (srcBLen - 1u); |
emh203 | 0:3d9c67d97d6f | 191 | py = pSrc1; |
emh203 | 0:3d9c67d97d6f | 192 | |
emh203 | 0:3d9c67d97d6f | 193 | /* ------------------------ |
emh203 | 0:3d9c67d97d6f | 194 | * Stage1 process |
emh203 | 0:3d9c67d97d6f | 195 | * ----------------------*/ |
emh203 | 0:3d9c67d97d6f | 196 | |
emh203 | 0:3d9c67d97d6f | 197 | /* The first stage starts here */ |
emh203 | 0:3d9c67d97d6f | 198 | while(blockSize1 > 0u) |
emh203 | 0:3d9c67d97d6f | 199 | { |
emh203 | 0:3d9c67d97d6f | 200 | /* Accumulator is made zero for every iteration */ |
emh203 | 0:3d9c67d97d6f | 201 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 202 | |
emh203 | 0:3d9c67d97d6f | 203 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emh203 | 0:3d9c67d97d6f | 204 | k = count >> 2; |
emh203 | 0:3d9c67d97d6f | 205 | |
emh203 | 0:3d9c67d97d6f | 206 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emh203 | 0:3d9c67d97d6f | 207 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 208 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 209 | { |
emh203 | 0:3d9c67d97d6f | 210 | /* x[0] , x[1] */ |
emh203 | 0:3d9c67d97d6f | 211 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 212 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 213 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 214 | |
emh203 | 0:3d9c67d97d6f | 215 | /* y[srcBLen - 4] , y[srcBLen - 3] */ |
emh203 | 0:3d9c67d97d6f | 216 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 217 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 218 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 219 | |
emh203 | 0:3d9c67d97d6f | 220 | /* x[0] * y[srcBLen - 4] */ |
emh203 | 0:3d9c67d97d6f | 221 | /* x[1] * y[srcBLen - 3] */ |
emh203 | 0:3d9c67d97d6f | 222 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 223 | |
emh203 | 0:3d9c67d97d6f | 224 | /* x[2] , x[3] */ |
emh203 | 0:3d9c67d97d6f | 225 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 226 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 227 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 228 | |
emh203 | 0:3d9c67d97d6f | 229 | /* y[srcBLen - 2] , y[srcBLen - 1] */ |
emh203 | 0:3d9c67d97d6f | 230 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 231 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 232 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 233 | |
emh203 | 0:3d9c67d97d6f | 234 | /* x[2] * y[srcBLen - 2] */ |
emh203 | 0:3d9c67d97d6f | 235 | /* x[3] * y[srcBLen - 1] */ |
emh203 | 0:3d9c67d97d6f | 236 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 237 | |
emh203 | 0:3d9c67d97d6f | 238 | |
emh203 | 0:3d9c67d97d6f | 239 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 240 | k--; |
emh203 | 0:3d9c67d97d6f | 241 | } |
emh203 | 0:3d9c67d97d6f | 242 | |
emh203 | 0:3d9c67d97d6f | 243 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
emh203 | 0:3d9c67d97d6f | 244 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 245 | k = count % 0x4u; |
emh203 | 0:3d9c67d97d6f | 246 | |
emh203 | 0:3d9c67d97d6f | 247 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 248 | { |
emh203 | 0:3d9c67d97d6f | 249 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 250 | /* x[0] * y[srcBLen - 1] */ |
emh203 | 0:3d9c67d97d6f | 251 | sum += (q31_t) ((q15_t) * px++ * *py++); |
emh203 | 0:3d9c67d97d6f | 252 | |
emh203 | 0:3d9c67d97d6f | 253 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 254 | k--; |
emh203 | 0:3d9c67d97d6f | 255 | } |
emh203 | 0:3d9c67d97d6f | 256 | |
emh203 | 0:3d9c67d97d6f | 257 | /* Store the result in the accumulator in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 258 | *pOut = (q7_t) (__SSAT(sum >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 259 | /* Destination pointer is updated according to the address modifier, inc */ |
emh203 | 0:3d9c67d97d6f | 260 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 261 | |
emh203 | 0:3d9c67d97d6f | 262 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emh203 | 0:3d9c67d97d6f | 263 | py = pSrc1 - count; |
emh203 | 0:3d9c67d97d6f | 264 | px = pIn1; |
emh203 | 0:3d9c67d97d6f | 265 | |
emh203 | 0:3d9c67d97d6f | 266 | /* Increment the MAC count */ |
emh203 | 0:3d9c67d97d6f | 267 | count++; |
emh203 | 0:3d9c67d97d6f | 268 | |
emh203 | 0:3d9c67d97d6f | 269 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 270 | blockSize1--; |
emh203 | 0:3d9c67d97d6f | 271 | } |
emh203 | 0:3d9c67d97d6f | 272 | |
emh203 | 0:3d9c67d97d6f | 273 | /* -------------------------- |
emh203 | 0:3d9c67d97d6f | 274 | * Initializations of stage2 |
emh203 | 0:3d9c67d97d6f | 275 | * ------------------------*/ |
emh203 | 0:3d9c67d97d6f | 276 | |
emh203 | 0:3d9c67d97d6f | 277 | /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] |
emh203 | 0:3d9c67d97d6f | 278 | * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] |
emh203 | 0:3d9c67d97d6f | 279 | * .... |
emh203 | 0:3d9c67d97d6f | 280 | * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emh203 | 0:3d9c67d97d6f | 281 | */ |
emh203 | 0:3d9c67d97d6f | 282 | |
emh203 | 0:3d9c67d97d6f | 283 | /* Working pointer of inputA */ |
emh203 | 0:3d9c67d97d6f | 284 | px = pIn1; |
emh203 | 0:3d9c67d97d6f | 285 | |
emh203 | 0:3d9c67d97d6f | 286 | /* Working pointer of inputB */ |
emh203 | 0:3d9c67d97d6f | 287 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 288 | |
emh203 | 0:3d9c67d97d6f | 289 | /* count is index by which the pointer pIn1 to be incremented */ |
emh203 | 0:3d9c67d97d6f | 290 | count = 0u; |
emh203 | 0:3d9c67d97d6f | 291 | |
emh203 | 0:3d9c67d97d6f | 292 | /* ------------------- |
emh203 | 0:3d9c67d97d6f | 293 | * Stage2 process |
emh203 | 0:3d9c67d97d6f | 294 | * ------------------*/ |
emh203 | 0:3d9c67d97d6f | 295 | |
emh203 | 0:3d9c67d97d6f | 296 | /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. |
emh203 | 0:3d9c67d97d6f | 297 | * So, to loop unroll over blockSize2, |
emh203 | 0:3d9c67d97d6f | 298 | * srcBLen should be greater than or equal to 4 */ |
emh203 | 0:3d9c67d97d6f | 299 | if(srcBLen >= 4u) |
emh203 | 0:3d9c67d97d6f | 300 | { |
emh203 | 0:3d9c67d97d6f | 301 | /* Loop unroll over blockSize2, by 4 */ |
emh203 | 0:3d9c67d97d6f | 302 | blkCnt = blockSize2 >> 2u; |
emh203 | 0:3d9c67d97d6f | 303 | |
emh203 | 0:3d9c67d97d6f | 304 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 305 | { |
emh203 | 0:3d9c67d97d6f | 306 | /* Set all accumulators to zero */ |
emh203 | 0:3d9c67d97d6f | 307 | acc0 = 0; |
emh203 | 0:3d9c67d97d6f | 308 | acc1 = 0; |
emh203 | 0:3d9c67d97d6f | 309 | acc2 = 0; |
emh203 | 0:3d9c67d97d6f | 310 | acc3 = 0; |
emh203 | 0:3d9c67d97d6f | 311 | |
emh203 | 0:3d9c67d97d6f | 312 | /* read x[0], x[1], x[2] samples */ |
emh203 | 0:3d9c67d97d6f | 313 | x0 = *px++; |
emh203 | 0:3d9c67d97d6f | 314 | x1 = *px++; |
emh203 | 0:3d9c67d97d6f | 315 | x2 = *px++; |
emh203 | 0:3d9c67d97d6f | 316 | |
emh203 | 0:3d9c67d97d6f | 317 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emh203 | 0:3d9c67d97d6f | 318 | k = srcBLen >> 2u; |
emh203 | 0:3d9c67d97d6f | 319 | |
emh203 | 0:3d9c67d97d6f | 320 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emh203 | 0:3d9c67d97d6f | 321 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 322 | do |
emh203 | 0:3d9c67d97d6f | 323 | { |
emh203 | 0:3d9c67d97d6f | 324 | /* Read y[0] sample */ |
emh203 | 0:3d9c67d97d6f | 325 | c0 = *py++; |
emh203 | 0:3d9c67d97d6f | 326 | /* Read y[1] sample */ |
emh203 | 0:3d9c67d97d6f | 327 | c1 = *py++; |
emh203 | 0:3d9c67d97d6f | 328 | |
emh203 | 0:3d9c67d97d6f | 329 | /* Read x[3] sample */ |
emh203 | 0:3d9c67d97d6f | 330 | x3 = *px++; |
emh203 | 0:3d9c67d97d6f | 331 | |
emh203 | 0:3d9c67d97d6f | 332 | /* x[0] and x[1] are packed */ |
emh203 | 0:3d9c67d97d6f | 333 | in1 = (q15_t) x0; |
emh203 | 0:3d9c67d97d6f | 334 | in2 = (q15_t) x1; |
emh203 | 0:3d9c67d97d6f | 335 | |
emh203 | 0:3d9c67d97d6f | 336 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 337 | |
emh203 | 0:3d9c67d97d6f | 338 | /* y[0] and y[1] are packed */ |
emh203 | 0:3d9c67d97d6f | 339 | in1 = (q15_t) c0; |
emh203 | 0:3d9c67d97d6f | 340 | in2 = (q15_t) c1; |
emh203 | 0:3d9c67d97d6f | 341 | |
emh203 | 0:3d9c67d97d6f | 342 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 343 | |
emh203 | 0:3d9c67d97d6f | 344 | /* acc0 += x[0] * y[0] + x[1] * y[1] */ |
emh203 | 0:3d9c67d97d6f | 345 | acc0 = __SMLAD(input1, input2, acc0); |
emh203 | 0:3d9c67d97d6f | 346 | |
emh203 | 0:3d9c67d97d6f | 347 | /* x[1] and x[2] are packed */ |
emh203 | 0:3d9c67d97d6f | 348 | in1 = (q15_t) x1; |
emh203 | 0:3d9c67d97d6f | 349 | in2 = (q15_t) x2; |
emh203 | 0:3d9c67d97d6f | 350 | |
emh203 | 0:3d9c67d97d6f | 351 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 352 | |
emh203 | 0:3d9c67d97d6f | 353 | /* acc1 += x[1] * y[0] + x[2] * y[1] */ |
emh203 | 0:3d9c67d97d6f | 354 | acc1 = __SMLAD(input1, input2, acc1); |
emh203 | 0:3d9c67d97d6f | 355 | |
emh203 | 0:3d9c67d97d6f | 356 | /* x[2] and x[3] are packed */ |
emh203 | 0:3d9c67d97d6f | 357 | in1 = (q15_t) x2; |
emh203 | 0:3d9c67d97d6f | 358 | in2 = (q15_t) x3; |
emh203 | 0:3d9c67d97d6f | 359 | |
emh203 | 0:3d9c67d97d6f | 360 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 361 | |
emh203 | 0:3d9c67d97d6f | 362 | /* acc2 += x[2] * y[0] + x[3] * y[1] */ |
emh203 | 0:3d9c67d97d6f | 363 | acc2 = __SMLAD(input1, input2, acc2); |
emh203 | 0:3d9c67d97d6f | 364 | |
emh203 | 0:3d9c67d97d6f | 365 | /* Read x[4] sample */ |
emh203 | 0:3d9c67d97d6f | 366 | x0 = *(px++); |
emh203 | 0:3d9c67d97d6f | 367 | |
emh203 | 0:3d9c67d97d6f | 368 | /* x[3] and x[4] are packed */ |
emh203 | 0:3d9c67d97d6f | 369 | in1 = (q15_t) x3; |
emh203 | 0:3d9c67d97d6f | 370 | in2 = (q15_t) x0; |
emh203 | 0:3d9c67d97d6f | 371 | |
emh203 | 0:3d9c67d97d6f | 372 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 373 | |
emh203 | 0:3d9c67d97d6f | 374 | /* acc3 += x[3] * y[0] + x[4] * y[1] */ |
emh203 | 0:3d9c67d97d6f | 375 | acc3 = __SMLAD(input1, input2, acc3); |
emh203 | 0:3d9c67d97d6f | 376 | |
emh203 | 0:3d9c67d97d6f | 377 | /* Read y[2] sample */ |
emh203 | 0:3d9c67d97d6f | 378 | c0 = *py++; |
emh203 | 0:3d9c67d97d6f | 379 | /* Read y[3] sample */ |
emh203 | 0:3d9c67d97d6f | 380 | c1 = *py++; |
emh203 | 0:3d9c67d97d6f | 381 | |
emh203 | 0:3d9c67d97d6f | 382 | /* Read x[5] sample */ |
emh203 | 0:3d9c67d97d6f | 383 | x1 = *px++; |
emh203 | 0:3d9c67d97d6f | 384 | |
emh203 | 0:3d9c67d97d6f | 385 | /* x[2] and x[3] are packed */ |
emh203 | 0:3d9c67d97d6f | 386 | in1 = (q15_t) x2; |
emh203 | 0:3d9c67d97d6f | 387 | in2 = (q15_t) x3; |
emh203 | 0:3d9c67d97d6f | 388 | |
emh203 | 0:3d9c67d97d6f | 389 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 390 | |
emh203 | 0:3d9c67d97d6f | 391 | /* y[2] and y[3] are packed */ |
emh203 | 0:3d9c67d97d6f | 392 | in1 = (q15_t) c0; |
emh203 | 0:3d9c67d97d6f | 393 | in2 = (q15_t) c1; |
emh203 | 0:3d9c67d97d6f | 394 | |
emh203 | 0:3d9c67d97d6f | 395 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 396 | |
emh203 | 0:3d9c67d97d6f | 397 | /* acc0 += x[2] * y[2] + x[3] * y[3] */ |
emh203 | 0:3d9c67d97d6f | 398 | acc0 = __SMLAD(input1, input2, acc0); |
emh203 | 0:3d9c67d97d6f | 399 | |
emh203 | 0:3d9c67d97d6f | 400 | /* x[3] and x[4] are packed */ |
emh203 | 0:3d9c67d97d6f | 401 | in1 = (q15_t) x3; |
emh203 | 0:3d9c67d97d6f | 402 | in2 = (q15_t) x0; |
emh203 | 0:3d9c67d97d6f | 403 | |
emh203 | 0:3d9c67d97d6f | 404 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 405 | |
emh203 | 0:3d9c67d97d6f | 406 | /* acc1 += x[3] * y[2] + x[4] * y[3] */ |
emh203 | 0:3d9c67d97d6f | 407 | acc1 = __SMLAD(input1, input2, acc1); |
emh203 | 0:3d9c67d97d6f | 408 | |
emh203 | 0:3d9c67d97d6f | 409 | /* x[4] and x[5] are packed */ |
emh203 | 0:3d9c67d97d6f | 410 | in1 = (q15_t) x0; |
emh203 | 0:3d9c67d97d6f | 411 | in2 = (q15_t) x1; |
emh203 | 0:3d9c67d97d6f | 412 | |
emh203 | 0:3d9c67d97d6f | 413 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 414 | |
emh203 | 0:3d9c67d97d6f | 415 | /* acc2 += x[4] * y[2] + x[5] * y[3] */ |
emh203 | 0:3d9c67d97d6f | 416 | acc2 = __SMLAD(input1, input2, acc2); |
emh203 | 0:3d9c67d97d6f | 417 | |
emh203 | 0:3d9c67d97d6f | 418 | /* Read x[6] sample */ |
emh203 | 0:3d9c67d97d6f | 419 | x2 = *px++; |
emh203 | 0:3d9c67d97d6f | 420 | |
emh203 | 0:3d9c67d97d6f | 421 | /* x[5] and x[6] are packed */ |
emh203 | 0:3d9c67d97d6f | 422 | in1 = (q15_t) x1; |
emh203 | 0:3d9c67d97d6f | 423 | in2 = (q15_t) x2; |
emh203 | 0:3d9c67d97d6f | 424 | |
emh203 | 0:3d9c67d97d6f | 425 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 426 | |
emh203 | 0:3d9c67d97d6f | 427 | /* acc3 += x[5] * y[2] + x[6] * y[3] */ |
emh203 | 0:3d9c67d97d6f | 428 | acc3 = __SMLAD(input1, input2, acc3); |
emh203 | 0:3d9c67d97d6f | 429 | |
emh203 | 0:3d9c67d97d6f | 430 | } while(--k); |
emh203 | 0:3d9c67d97d6f | 431 | |
emh203 | 0:3d9c67d97d6f | 432 | /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. |
emh203 | 0:3d9c67d97d6f | 433 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 434 | k = srcBLen % 0x4u; |
emh203 | 0:3d9c67d97d6f | 435 | |
emh203 | 0:3d9c67d97d6f | 436 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 437 | { |
emh203 | 0:3d9c67d97d6f | 438 | /* Read y[4] sample */ |
emh203 | 0:3d9c67d97d6f | 439 | c0 = *py++; |
emh203 | 0:3d9c67d97d6f | 440 | |
emh203 | 0:3d9c67d97d6f | 441 | /* Read x[7] sample */ |
emh203 | 0:3d9c67d97d6f | 442 | x3 = *px++; |
emh203 | 0:3d9c67d97d6f | 443 | |
emh203 | 0:3d9c67d97d6f | 444 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 445 | /* acc0 += x[4] * y[4] */ |
emh203 | 0:3d9c67d97d6f | 446 | acc0 += ((q15_t) x0 * c0); |
emh203 | 0:3d9c67d97d6f | 447 | /* acc1 += x[5] * y[4] */ |
emh203 | 0:3d9c67d97d6f | 448 | acc1 += ((q15_t) x1 * c0); |
emh203 | 0:3d9c67d97d6f | 449 | /* acc2 += x[6] * y[4] */ |
emh203 | 0:3d9c67d97d6f | 450 | acc2 += ((q15_t) x2 * c0); |
emh203 | 0:3d9c67d97d6f | 451 | /* acc3 += x[7] * y[4] */ |
emh203 | 0:3d9c67d97d6f | 452 | acc3 += ((q15_t) x3 * c0); |
emh203 | 0:3d9c67d97d6f | 453 | |
emh203 | 0:3d9c67d97d6f | 454 | /* Reuse the present samples for the next MAC */ |
emh203 | 0:3d9c67d97d6f | 455 | x0 = x1; |
emh203 | 0:3d9c67d97d6f | 456 | x1 = x2; |
emh203 | 0:3d9c67d97d6f | 457 | x2 = x3; |
emh203 | 0:3d9c67d97d6f | 458 | |
emh203 | 0:3d9c67d97d6f | 459 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 460 | k--; |
emh203 | 0:3d9c67d97d6f | 461 | } |
emh203 | 0:3d9c67d97d6f | 462 | |
emh203 | 0:3d9c67d97d6f | 463 | /* Store the result in the accumulator in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 464 | *pOut = (q7_t) (__SSAT(acc0 >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 465 | /* Destination pointer is updated according to the address modifier, inc */ |
emh203 | 0:3d9c67d97d6f | 466 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 467 | |
emh203 | 0:3d9c67d97d6f | 468 | *pOut = (q7_t) (__SSAT(acc1 >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 469 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 470 | |
emh203 | 0:3d9c67d97d6f | 471 | *pOut = (q7_t) (__SSAT(acc2 >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 472 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 473 | |
emh203 | 0:3d9c67d97d6f | 474 | *pOut = (q7_t) (__SSAT(acc3 >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 475 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 476 | |
emh203 | 0:3d9c67d97d6f | 477 | count += 4u; |
emh203 | 0:3d9c67d97d6f | 478 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emh203 | 0:3d9c67d97d6f | 479 | px = pIn1 + count; |
emh203 | 0:3d9c67d97d6f | 480 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 481 | |
emh203 | 0:3d9c67d97d6f | 482 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 483 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 484 | } |
emh203 | 0:3d9c67d97d6f | 485 | |
emh203 | 0:3d9c67d97d6f | 486 | /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. |
emh203 | 0:3d9c67d97d6f | 487 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 488 | blkCnt = blockSize2 % 0x4u; |
emh203 | 0:3d9c67d97d6f | 489 | |
emh203 | 0:3d9c67d97d6f | 490 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 491 | { |
emh203 | 0:3d9c67d97d6f | 492 | /* Accumulator is made zero for every iteration */ |
emh203 | 0:3d9c67d97d6f | 493 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 494 | |
emh203 | 0:3d9c67d97d6f | 495 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emh203 | 0:3d9c67d97d6f | 496 | k = srcBLen >> 2u; |
emh203 | 0:3d9c67d97d6f | 497 | |
emh203 | 0:3d9c67d97d6f | 498 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emh203 | 0:3d9c67d97d6f | 499 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 500 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 501 | { |
emh203 | 0:3d9c67d97d6f | 502 | /* Reading two inputs of SrcA buffer and packing */ |
emh203 | 0:3d9c67d97d6f | 503 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 504 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 505 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 506 | |
emh203 | 0:3d9c67d97d6f | 507 | /* Reading two inputs of SrcB buffer and packing */ |
emh203 | 0:3d9c67d97d6f | 508 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 509 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 510 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 511 | |
emh203 | 0:3d9c67d97d6f | 512 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 513 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 514 | |
emh203 | 0:3d9c67d97d6f | 515 | /* Reading two inputs of SrcA buffer and packing */ |
emh203 | 0:3d9c67d97d6f | 516 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 517 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 518 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 519 | |
emh203 | 0:3d9c67d97d6f | 520 | /* Reading two inputs of SrcB buffer and packing */ |
emh203 | 0:3d9c67d97d6f | 521 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 522 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 523 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 524 | |
emh203 | 0:3d9c67d97d6f | 525 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 526 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 527 | |
emh203 | 0:3d9c67d97d6f | 528 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 529 | k--; |
emh203 | 0:3d9c67d97d6f | 530 | } |
emh203 | 0:3d9c67d97d6f | 531 | |
emh203 | 0:3d9c67d97d6f | 532 | /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. |
emh203 | 0:3d9c67d97d6f | 533 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 534 | k = srcBLen % 0x4u; |
emh203 | 0:3d9c67d97d6f | 535 | |
emh203 | 0:3d9c67d97d6f | 536 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 537 | { |
emh203 | 0:3d9c67d97d6f | 538 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 539 | sum += ((q15_t) * px++ * *py++); |
emh203 | 0:3d9c67d97d6f | 540 | |
emh203 | 0:3d9c67d97d6f | 541 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 542 | k--; |
emh203 | 0:3d9c67d97d6f | 543 | } |
emh203 | 0:3d9c67d97d6f | 544 | |
emh203 | 0:3d9c67d97d6f | 545 | /* Store the result in the accumulator in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 546 | *pOut = (q7_t) (__SSAT(sum >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 547 | /* Destination pointer is updated according to the address modifier, inc */ |
emh203 | 0:3d9c67d97d6f | 548 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 549 | |
emh203 | 0:3d9c67d97d6f | 550 | /* Increment the pointer pIn1 index, count by 1 */ |
emh203 | 0:3d9c67d97d6f | 551 | count++; |
emh203 | 0:3d9c67d97d6f | 552 | |
emh203 | 0:3d9c67d97d6f | 553 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emh203 | 0:3d9c67d97d6f | 554 | px = pIn1 + count; |
emh203 | 0:3d9c67d97d6f | 555 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 556 | |
emh203 | 0:3d9c67d97d6f | 557 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 558 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 559 | } |
emh203 | 0:3d9c67d97d6f | 560 | } |
emh203 | 0:3d9c67d97d6f | 561 | else |
emh203 | 0:3d9c67d97d6f | 562 | { |
emh203 | 0:3d9c67d97d6f | 563 | /* If the srcBLen is not a multiple of 4, |
emh203 | 0:3d9c67d97d6f | 564 | * the blockSize2 loop cannot be unrolled by 4 */ |
emh203 | 0:3d9c67d97d6f | 565 | blkCnt = blockSize2; |
emh203 | 0:3d9c67d97d6f | 566 | |
emh203 | 0:3d9c67d97d6f | 567 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 568 | { |
emh203 | 0:3d9c67d97d6f | 569 | /* Accumulator is made zero for every iteration */ |
emh203 | 0:3d9c67d97d6f | 570 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 571 | |
emh203 | 0:3d9c67d97d6f | 572 | /* Loop over srcBLen */ |
emh203 | 0:3d9c67d97d6f | 573 | k = srcBLen; |
emh203 | 0:3d9c67d97d6f | 574 | |
emh203 | 0:3d9c67d97d6f | 575 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 576 | { |
emh203 | 0:3d9c67d97d6f | 577 | /* Perform the multiply-accumulate */ |
emh203 | 0:3d9c67d97d6f | 578 | sum += ((q15_t) * px++ * *py++); |
emh203 | 0:3d9c67d97d6f | 579 | |
emh203 | 0:3d9c67d97d6f | 580 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 581 | k--; |
emh203 | 0:3d9c67d97d6f | 582 | } |
emh203 | 0:3d9c67d97d6f | 583 | |
emh203 | 0:3d9c67d97d6f | 584 | /* Store the result in the accumulator in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 585 | *pOut = (q7_t) (__SSAT(sum >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 586 | /* Destination pointer is updated according to the address modifier, inc */ |
emh203 | 0:3d9c67d97d6f | 587 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 588 | |
emh203 | 0:3d9c67d97d6f | 589 | /* Increment the MAC count */ |
emh203 | 0:3d9c67d97d6f | 590 | count++; |
emh203 | 0:3d9c67d97d6f | 591 | |
emh203 | 0:3d9c67d97d6f | 592 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emh203 | 0:3d9c67d97d6f | 593 | px = pIn1 + count; |
emh203 | 0:3d9c67d97d6f | 594 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 595 | |
emh203 | 0:3d9c67d97d6f | 596 | |
emh203 | 0:3d9c67d97d6f | 597 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 598 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 599 | } |
emh203 | 0:3d9c67d97d6f | 600 | } |
emh203 | 0:3d9c67d97d6f | 601 | |
emh203 | 0:3d9c67d97d6f | 602 | /* -------------------------- |
emh203 | 0:3d9c67d97d6f | 603 | * Initializations of stage3 |
emh203 | 0:3d9c67d97d6f | 604 | * -------------------------*/ |
emh203 | 0:3d9c67d97d6f | 605 | |
emh203 | 0:3d9c67d97d6f | 606 | /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emh203 | 0:3d9c67d97d6f | 607 | * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] |
emh203 | 0:3d9c67d97d6f | 608 | * .... |
emh203 | 0:3d9c67d97d6f | 609 | * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] |
emh203 | 0:3d9c67d97d6f | 610 | * sum += x[srcALen-1] * y[0] |
emh203 | 0:3d9c67d97d6f | 611 | */ |
emh203 | 0:3d9c67d97d6f | 612 | |
emh203 | 0:3d9c67d97d6f | 613 | /* In this stage the MAC operations are decreased by 1 for every iteration. |
emh203 | 0:3d9c67d97d6f | 614 | The count variable holds the number of MAC operations performed */ |
emh203 | 0:3d9c67d97d6f | 615 | count = srcBLen - 1u; |
emh203 | 0:3d9c67d97d6f | 616 | |
emh203 | 0:3d9c67d97d6f | 617 | /* Working pointer of inputA */ |
emh203 | 0:3d9c67d97d6f | 618 | pSrc1 = pIn1 + (srcALen - (srcBLen - 1u)); |
emh203 | 0:3d9c67d97d6f | 619 | px = pSrc1; |
emh203 | 0:3d9c67d97d6f | 620 | |
emh203 | 0:3d9c67d97d6f | 621 | /* Working pointer of inputB */ |
emh203 | 0:3d9c67d97d6f | 622 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 623 | |
emh203 | 0:3d9c67d97d6f | 624 | /* ------------------- |
emh203 | 0:3d9c67d97d6f | 625 | * Stage3 process |
emh203 | 0:3d9c67d97d6f | 626 | * ------------------*/ |
emh203 | 0:3d9c67d97d6f | 627 | |
emh203 | 0:3d9c67d97d6f | 628 | while(blockSize3 > 0u) |
emh203 | 0:3d9c67d97d6f | 629 | { |
emh203 | 0:3d9c67d97d6f | 630 | /* Accumulator is made zero for every iteration */ |
emh203 | 0:3d9c67d97d6f | 631 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 632 | |
emh203 | 0:3d9c67d97d6f | 633 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
emh203 | 0:3d9c67d97d6f | 634 | k = count >> 2u; |
emh203 | 0:3d9c67d97d6f | 635 | |
emh203 | 0:3d9c67d97d6f | 636 | /* First part of the processing with loop unrolling. Compute 4 MACs at a time. |
emh203 | 0:3d9c67d97d6f | 637 | ** a second loop below computes MACs for the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 638 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 639 | { |
emh203 | 0:3d9c67d97d6f | 640 | /* x[srcALen - srcBLen + 1] , x[srcALen - srcBLen + 2] */ |
emh203 | 0:3d9c67d97d6f | 641 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 642 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 643 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 644 | |
emh203 | 0:3d9c67d97d6f | 645 | /* y[0] , y[1] */ |
emh203 | 0:3d9c67d97d6f | 646 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 647 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 648 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 649 | |
emh203 | 0:3d9c67d97d6f | 650 | /* sum += x[srcALen - srcBLen + 1] * y[0] */ |
emh203 | 0:3d9c67d97d6f | 651 | /* sum += x[srcALen - srcBLen + 2] * y[1] */ |
emh203 | 0:3d9c67d97d6f | 652 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 653 | |
emh203 | 0:3d9c67d97d6f | 654 | /* x[srcALen - srcBLen + 3] , x[srcALen - srcBLen + 4] */ |
emh203 | 0:3d9c67d97d6f | 655 | in1 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 656 | in2 = (q15_t) * px++; |
emh203 | 0:3d9c67d97d6f | 657 | input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 658 | |
emh203 | 0:3d9c67d97d6f | 659 | /* y[2] , y[3] */ |
emh203 | 0:3d9c67d97d6f | 660 | in1 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 661 | in2 = (q15_t) * py++; |
emh203 | 0:3d9c67d97d6f | 662 | input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16); |
emh203 | 0:3d9c67d97d6f | 663 | |
emh203 | 0:3d9c67d97d6f | 664 | /* sum += x[srcALen - srcBLen + 3] * y[2] */ |
emh203 | 0:3d9c67d97d6f | 665 | /* sum += x[srcALen - srcBLen + 4] * y[3] */ |
emh203 | 0:3d9c67d97d6f | 666 | sum = __SMLAD(input1, input2, sum); |
emh203 | 0:3d9c67d97d6f | 667 | |
emh203 | 0:3d9c67d97d6f | 668 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 669 | k--; |
emh203 | 0:3d9c67d97d6f | 670 | } |
emh203 | 0:3d9c67d97d6f | 671 | |
emh203 | 0:3d9c67d97d6f | 672 | /* If the count is not a multiple of 4, compute any remaining MACs here. |
emh203 | 0:3d9c67d97d6f | 673 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 674 | k = count % 0x4u; |
emh203 | 0:3d9c67d97d6f | 675 | |
emh203 | 0:3d9c67d97d6f | 676 | while(k > 0u) |
emh203 | 0:3d9c67d97d6f | 677 | { |
emh203 | 0:3d9c67d97d6f | 678 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 679 | sum += ((q15_t) * px++ * *py++); |
emh203 | 0:3d9c67d97d6f | 680 | |
emh203 | 0:3d9c67d97d6f | 681 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 682 | k--; |
emh203 | 0:3d9c67d97d6f | 683 | } |
emh203 | 0:3d9c67d97d6f | 684 | |
emh203 | 0:3d9c67d97d6f | 685 | /* Store the result in the accumulator in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 686 | *pOut = (q7_t) (__SSAT(sum >> 7, 8)); |
emh203 | 0:3d9c67d97d6f | 687 | /* Destination pointer is updated according to the address modifier, inc */ |
emh203 | 0:3d9c67d97d6f | 688 | pOut += inc; |
emh203 | 0:3d9c67d97d6f | 689 | |
emh203 | 0:3d9c67d97d6f | 690 | /* Update the inputA and inputB pointers for next MAC calculation */ |
emh203 | 0:3d9c67d97d6f | 691 | px = ++pSrc1; |
emh203 | 0:3d9c67d97d6f | 692 | py = pIn2; |
emh203 | 0:3d9c67d97d6f | 693 | |
emh203 | 0:3d9c67d97d6f | 694 | /* Decrement the MAC count */ |
emh203 | 0:3d9c67d97d6f | 695 | count--; |
emh203 | 0:3d9c67d97d6f | 696 | |
emh203 | 0:3d9c67d97d6f | 697 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 698 | blockSize3--; |
emh203 | 0:3d9c67d97d6f | 699 | } |
emh203 | 0:3d9c67d97d6f | 700 | |
emh203 | 0:3d9c67d97d6f | 701 | #else |
emh203 | 0:3d9c67d97d6f | 702 | |
emh203 | 0:3d9c67d97d6f | 703 | /* Run the below code for Cortex-M0 */ |
emh203 | 0:3d9c67d97d6f | 704 | |
emh203 | 0:3d9c67d97d6f | 705 | q7_t *pIn1 = pSrcA; /* inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 706 | q7_t *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 707 | q31_t sum; /* Accumulator */ |
emh203 | 0:3d9c67d97d6f | 708 | uint32_t i = 0u, j; /* loop counters */ |
emh203 | 0:3d9c67d97d6f | 709 | uint32_t inv = 0u; /* Reverse order flag */ |
emh203 | 0:3d9c67d97d6f | 710 | uint32_t tot = 0u; /* Length */ |
emh203 | 0:3d9c67d97d6f | 711 | |
emh203 | 0:3d9c67d97d6f | 712 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emh203 | 0:3d9c67d97d6f | 713 | /* srcB is always made to slide across srcA. */ |
emh203 | 0:3d9c67d97d6f | 714 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emh203 | 0:3d9c67d97d6f | 715 | /* But CORR(x, y) is reverse of CORR(y, x) */ |
emh203 | 0:3d9c67d97d6f | 716 | /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 717 | /* and a varaible, inv is set to 1 */ |
emh203 | 0:3d9c67d97d6f | 718 | /* If lengths are not equal then zero pad has to be done to make the two |
emh203 | 0:3d9c67d97d6f | 719 | * inputs of same length. But to improve the performance, we include zeroes |
emh203 | 0:3d9c67d97d6f | 720 | * in the output instead of zero padding either of the the inputs*/ |
emh203 | 0:3d9c67d97d6f | 721 | /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the |
emh203 | 0:3d9c67d97d6f | 722 | * starting of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 723 | /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the |
emh203 | 0:3d9c67d97d6f | 724 | * ending of the output buffer */ |
emh203 | 0:3d9c67d97d6f | 725 | /* Once the zero padding is done the remaining of the output is calcualted |
emh203 | 0:3d9c67d97d6f | 726 | * using convolution but with the shorter signal time shifted. */ |
emh203 | 0:3d9c67d97d6f | 727 | |
emh203 | 0:3d9c67d97d6f | 728 | /* Calculate the length of the remaining sequence */ |
emh203 | 0:3d9c67d97d6f | 729 | tot = ((srcALen + srcBLen) - 2u); |
emh203 | 0:3d9c67d97d6f | 730 | |
emh203 | 0:3d9c67d97d6f | 731 | if(srcALen > srcBLen) |
emh203 | 0:3d9c67d97d6f | 732 | { |
emh203 | 0:3d9c67d97d6f | 733 | /* Calculating the number of zeros to be padded to the output */ |
emh203 | 0:3d9c67d97d6f | 734 | j = srcALen - srcBLen; |
emh203 | 0:3d9c67d97d6f | 735 | |
emh203 | 0:3d9c67d97d6f | 736 | /* Initialise the pointer after zero padding */ |
emh203 | 0:3d9c67d97d6f | 737 | pDst += j; |
emh203 | 0:3d9c67d97d6f | 738 | } |
emh203 | 0:3d9c67d97d6f | 739 | |
emh203 | 0:3d9c67d97d6f | 740 | else if(srcALen < srcBLen) |
emh203 | 0:3d9c67d97d6f | 741 | { |
emh203 | 0:3d9c67d97d6f | 742 | /* Initialization to inputB pointer */ |
emh203 | 0:3d9c67d97d6f | 743 | pIn1 = pSrcB; |
emh203 | 0:3d9c67d97d6f | 744 | |
emh203 | 0:3d9c67d97d6f | 745 | /* Initialization to the end of inputA pointer */ |
emh203 | 0:3d9c67d97d6f | 746 | pIn2 = pSrcA + (srcALen - 1u); |
emh203 | 0:3d9c67d97d6f | 747 | |
emh203 | 0:3d9c67d97d6f | 748 | /* Initialisation of the pointer after zero padding */ |
emh203 | 0:3d9c67d97d6f | 749 | pDst = pDst + tot; |
emh203 | 0:3d9c67d97d6f | 750 | |
emh203 | 0:3d9c67d97d6f | 751 | /* Swapping the lengths */ |
emh203 | 0:3d9c67d97d6f | 752 | j = srcALen; |
emh203 | 0:3d9c67d97d6f | 753 | srcALen = srcBLen; |
emh203 | 0:3d9c67d97d6f | 754 | srcBLen = j; |
emh203 | 0:3d9c67d97d6f | 755 | |
emh203 | 0:3d9c67d97d6f | 756 | /* Setting the reverse flag */ |
emh203 | 0:3d9c67d97d6f | 757 | inv = 1; |
emh203 | 0:3d9c67d97d6f | 758 | |
emh203 | 0:3d9c67d97d6f | 759 | } |
emh203 | 0:3d9c67d97d6f | 760 | |
emh203 | 0:3d9c67d97d6f | 761 | /* Loop to calculate convolution for output length number of times */ |
emh203 | 0:3d9c67d97d6f | 762 | for (i = 0u; i <= tot; i++) |
emh203 | 0:3d9c67d97d6f | 763 | { |
emh203 | 0:3d9c67d97d6f | 764 | /* Initialize sum with zero to carry on MAC operations */ |
emh203 | 0:3d9c67d97d6f | 765 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 766 | |
emh203 | 0:3d9c67d97d6f | 767 | /* Loop to perform MAC operations according to convolution equation */ |
emh203 | 0:3d9c67d97d6f | 768 | for (j = 0u; j <= i; j++) |
emh203 | 0:3d9c67d97d6f | 769 | { |
emh203 | 0:3d9c67d97d6f | 770 | /* Check the array limitations */ |
emh203 | 0:3d9c67d97d6f | 771 | if((((i - j) < srcBLen) && (j < srcALen))) |
emh203 | 0:3d9c67d97d6f | 772 | { |
emh203 | 0:3d9c67d97d6f | 773 | /* z[i] += x[i-j] * y[j] */ |
emh203 | 0:3d9c67d97d6f | 774 | sum += ((q15_t) pIn1[j] * pIn2[-((int32_t) i - j)]); |
emh203 | 0:3d9c67d97d6f | 775 | } |
emh203 | 0:3d9c67d97d6f | 776 | } |
emh203 | 0:3d9c67d97d6f | 777 | /* Store the output in the destination buffer */ |
emh203 | 0:3d9c67d97d6f | 778 | if(inv == 1) |
emh203 | 0:3d9c67d97d6f | 779 | *pDst-- = (q7_t) __SSAT((sum >> 7u), 8u); |
emh203 | 0:3d9c67d97d6f | 780 | else |
emh203 | 0:3d9c67d97d6f | 781 | *pDst++ = (q7_t) __SSAT((sum >> 7u), 8u); |
emh203 | 0:3d9c67d97d6f | 782 | } |
emh203 | 0:3d9c67d97d6f | 783 | |
emh203 | 0:3d9c67d97d6f | 784 | #endif /* #ifndef ARM_MATH_CM0_FAMILY */ |
emh203 | 0:3d9c67d97d6f | 785 | |
emh203 | 0:3d9c67d97d6f | 786 | } |
emh203 | 0:3d9c67d97d6f | 787 | |
emh203 | 0:3d9c67d97d6f | 788 | /** |
emh203 | 0:3d9c67d97d6f | 789 | * @} end of Corr group |
emh203 | 0:3d9c67d97d6f | 790 | */ |