| User | Revision | Line number | New contents of line |
|---|
| elt14lpo |
1:d0884279b41d
|
1
|
/* ----------------------------------------------------------------------------
|
| elt14lpo |
1:d0884279b41d
|
2
|
* Copyright (C) 2010 ARM Limited. All rights reserved.
|
| elt14lpo |
1:d0884279b41d
|
3
|
*
|
| elt14lpo |
1:d0884279b41d
|
4
|
* $Date: 29. November 2010
|
| elt14lpo |
1:d0884279b41d
|
5
|
* $Revision: V1.0.3
|
| elt14lpo |
1:d0884279b41d
|
6
|
*
|
| elt14lpo |
1:d0884279b41d
|
7
|
* Project: CMSIS DSP Library
|
| elt14lpo |
1:d0884279b41d
|
8
|
* Title: arm_correlate_f32.c
|
| elt14lpo |
1:d0884279b41d
|
9
|
*
|
| elt14lpo |
1:d0884279b41d
|
10
|
* Description: Correlation for floating-point sequences.
|
| elt14lpo |
1:d0884279b41d
|
11
|
*
|
| elt14lpo |
1:d0884279b41d
|
12
|
* Target Processor: Cortex-M4/Cortex-M3
|
| elt14lpo |
1:d0884279b41d
|
13
|
*
|
| elt14lpo |
1:d0884279b41d
|
14
|
* Version 1.0.3 2010/11/29
|
| elt14lpo |
1:d0884279b41d
|
15
|
* Re-organized the CMSIS folders and updated documentation.
|
| elt14lpo |
1:d0884279b41d
|
16
|
*
|
| elt14lpo |
1:d0884279b41d
|
17
|
* Version 1.0.2 2010/11/11
|
| elt14lpo |
1:d0884279b41d
|
18
|
* Documentation updated.
|
| elt14lpo |
1:d0884279b41d
|
19
|
*
|
| elt14lpo |
1:d0884279b41d
|
20
|
* Version 1.0.1 2010/10/05
|
| elt14lpo |
1:d0884279b41d
|
21
|
* Production release and review comments incorporated.
|
| elt14lpo |
1:d0884279b41d
|
22
|
*
|
| elt14lpo |
1:d0884279b41d
|
23
|
* Version 1.0.0 2010/09/20
|
| elt14lpo |
1:d0884279b41d
|
24
|
* Production release and review comments incorporated
|
| elt14lpo |
1:d0884279b41d
|
25
|
*
|
| elt14lpo |
1:d0884279b41d
|
26
|
* Version 0.0.7 2010/06/10
|
| elt14lpo |
1:d0884279b41d
|
27
|
* Misra-C changes done
|
| elt14lpo |
1:d0884279b41d
|
28
|
*
|
| elt14lpo |
1:d0884279b41d
|
29
|
* -------------------------------------------------------------------------- */
|
| elt14lpo |
1:d0884279b41d
|
30
|
|
| elt14lpo |
1:d0884279b41d
|
31
|
#include "arm_math.h"
|
| elt14lpo |
1:d0884279b41d
|
32
|
#include "arm_correlate_f32.h"
|
| elt14lpo |
1:d0884279b41d
|
33
|
|
| elt14lpo |
1:d0884279b41d
|
34
|
/**
|
| elt14lpo |
1:d0884279b41d
|
35
|
* @ingroup groupFilters
|
| elt14lpo |
1:d0884279b41d
|
36
|
*/
|
| elt14lpo |
1:d0884279b41d
|
37
|
|
| elt14lpo |
1:d0884279b41d
|
38
|
/**
|
| elt14lpo |
1:d0884279b41d
|
39
|
* @defgroup Corr Correlation
|
| elt14lpo |
1:d0884279b41d
|
40
|
*
|
| elt14lpo |
1:d0884279b41d
|
41
|
* Correlation is a mathematical operation that is similar to convolution.
|
| elt14lpo |
1:d0884279b41d
|
42
|
* As with convolution, correlation uses two signals to produce a third signal.
|
| elt14lpo |
1:d0884279b41d
|
43
|
* The underlying algorithms in correlation and convolution are identical except that one of the inputs is flipped in convolution.
|
| elt14lpo |
1:d0884279b41d
|
44
|
* Correlation is commonly used to measure the similarity between two signals.
|
| elt14lpo |
1:d0884279b41d
|
45
|
* It has applications in pattern recognition, cryptanalysis, and searching.
|
| elt14lpo |
1:d0884279b41d
|
46
|
* The CMSIS library provides correlation functions for Q7, Q15, Q31 and floating-point data types.
|
| elt14lpo |
1:d0884279b41d
|
47
|
* Fast versions of the Q15 and Q31 functions are also provided.
|
| elt14lpo |
1:d0884279b41d
|
48
|
*
|
| elt14lpo |
1:d0884279b41d
|
49
|
* \par Algorithm
|
| elt14lpo |
1:d0884279b41d
|
50
|
* Let <code>a[n]</code> and <code>b[n]</code> be sequences of length <code>srcALen</code> and <code>srcBLen</code> samples respectively.
|
| elt14lpo |
1:d0884279b41d
|
51
|
* The convolution of the two signals is denoted by
|
| elt14lpo |
1:d0884279b41d
|
52
|
* <pre>
|
| elt14lpo |
1:d0884279b41d
|
53
|
* c[n] = a[n] * b[n]
|
| elt14lpo |
1:d0884279b41d
|
54
|
* </pre>
|
| elt14lpo |
1:d0884279b41d
|
55
|
* In correlation, one of the signals is flipped in time
|
| elt14lpo |
1:d0884279b41d
|
56
|
* <pre>
|
| elt14lpo |
1:d0884279b41d
|
57
|
* c[n] = a[n] * b[-n]
|
| elt14lpo |
1:d0884279b41d
|
58
|
* </pre>
|
| elt14lpo |
1:d0884279b41d
|
59
|
*
|
| elt14lpo |
1:d0884279b41d
|
60
|
* \par
|
| elt14lpo |
1:d0884279b41d
|
61
|
* and this is mathematically defined as
|
| elt14lpo |
1:d0884279b41d
|
62
|
* \image html CorrelateEquation.gif
|
| elt14lpo |
1:d0884279b41d
|
63
|
* \par
|
| elt14lpo |
1:d0884279b41d
|
64
|
* 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 |
1:d0884279b41d
|
65
|
* 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 |
1:d0884279b41d
|
66
|
* 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 |
1:d0884279b41d
|
67
|
*
|
| elt14lpo |
1:d0884279b41d
|
68
|
* <b>Fixed-Point Behavior</b>
|
| elt14lpo |
1:d0884279b41d
|
69
|
* \par
|
| elt14lpo |
1:d0884279b41d
|
70
|
* Correlation requires summing up a large number of intermediate products.
|
| elt14lpo |
1:d0884279b41d
|
71
|
* As such, the Q7, Q15, and Q31 functions run a risk of overflow and saturation.
|
| elt14lpo |
1:d0884279b41d
|
72
|
* Refer to the function specific documentation below for further details of the particular algorithm used.
|
| elt14lpo |
1:d0884279b41d
|
73
|
*/
|
| elt14lpo |
1:d0884279b41d
|
74
|
|
| elt14lpo |
1:d0884279b41d
|
75
|
/**
|
| elt14lpo |
1:d0884279b41d
|
76
|
* @addtogroup Corr
|
| elt14lpo |
1:d0884279b41d
|
77
|
* @{
|
| elt14lpo |
1:d0884279b41d
|
78
|
*/
|
| elt14lpo |
1:d0884279b41d
|
79
|
/**
|
| elt14lpo |
1:d0884279b41d
|
80
|
* @brief Correlation of floating-point sequences
|
| elt14lpo |
1:d0884279b41d
|
81
|
* @param[in] *pSrcA points to the first input sequence.
|
| elt14lpo |
1:d0884279b41d
|
82
|
* @param[in] srcALen length of the first input sequence.
|
| elt14lpo |
1:d0884279b41d
|
83
|
* @param[in] *pSrcB points to the second input sequence.
|
| elt14lpo |
1:d0884279b41d
|
84
|
* @param[in] srcBLen length of the second input sequence.
|
| elt14lpo |
1:d0884279b41d
|
85
|
* @param[out] *pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1.
|
| elt14lpo |
1:d0884279b41d
|
86
|
* @return none.
|
| elt14lpo |
1:d0884279b41d
|
87
|
*/
|
| elt14lpo |
1:d0884279b41d
|
88
|
|
| elt14lpo |
1:d0884279b41d
|
89
|
void arm_correlate_f32(float32_t * pSrcA, uint32_t srcALen, float32_t * pSrcB, uint32_t srcBLen, float32_t * pDst){
|
| elt14lpo |
1:d0884279b41d
|
90
|
float32_t *pIn1; /* inputA pointer */
|
| elt14lpo |
1:d0884279b41d
|
91
|
float32_t *pIn2; /* inputB pointer */
|
| elt14lpo |
1:d0884279b41d
|
92
|
float32_t *pOut = pDst; /* output pointer */
|
| elt14lpo |
1:d0884279b41d
|
93
|
float32_t *px; /* Intermediate inputA pointer */
|
| elt14lpo |
1:d0884279b41d
|
94
|
float32_t *py; /* Intermediate inputB pointer */
|
| elt14lpo |
1:d0884279b41d
|
95
|
float32_t *pSrc1; /* Intermediate pointers */
|
| elt14lpo |
1:d0884279b41d
|
96
|
float32_t sum, acc0, acc1, acc2, acc3; /* Accumulators */
|
| elt14lpo |
1:d0884279b41d
|
97
|
float32_t x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */
|
| elt14lpo |
1:d0884279b41d
|
98
|
uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counters */
|
| elt14lpo |
1:d0884279b41d
|
99
|
int32_t inc = 1; /* Destination address modifier */
|
| elt14lpo |
1:d0884279b41d
|
100
|
|
| elt14lpo |
1:d0884279b41d
|
101
|
|
| elt14lpo |
1:d0884279b41d
|
102
|
/* The algorithm implementation is based on the lengths of the inputs. */
|
| elt14lpo |
1:d0884279b41d
|
103
|
/* srcB is always made to slide across srcA. */
|
| elt14lpo |
1:d0884279b41d
|
104
|
/* So srcBLen is always considered as shorter or equal to srcALen */
|
| elt14lpo |
1:d0884279b41d
|
105
|
/* But CORR(x, y) is reverse of CORR(y, x) */
|
| elt14lpo |
1:d0884279b41d
|
106
|
/* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
|
| elt14lpo |
1:d0884279b41d
|
107
|
/* and the destination pointer modifier, inc is set to -1 */
|
| elt14lpo |
1:d0884279b41d
|
108
|
/* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */
|
| elt14lpo |
1:d0884279b41d
|
109
|
/* But to improve the performance,
|
| elt14lpo |
1:d0884279b41d
|
110
|
* we include zeroes in the output instead of zero padding either of the the inputs*/
|
| elt14lpo |
1:d0884279b41d
|
111
|
/* If srcALen > srcBLen,
|
| elt14lpo |
1:d0884279b41d
|
112
|
* (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */
|
| elt14lpo |
1:d0884279b41d
|
113
|
/* If srcALen < srcBLen,
|
| elt14lpo |
1:d0884279b41d
|
114
|
* (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */
|
| elt14lpo |
1:d0884279b41d
|
115
|
if(srcALen >= srcBLen) {
|
| elt14lpo |
1:d0884279b41d
|
116
|
/* Initialization of inputA pointer */
|
| elt14lpo |
1:d0884279b41d
|
117
|
pIn1 = pSrcA;
|
| elt14lpo |
1:d0884279b41d
|
118
|
|
| elt14lpo |
1:d0884279b41d
|
119
|
/* Initialization of inputB pointer */
|
| elt14lpo |
1:d0884279b41d
|
120
|
pIn2 = pSrcB;
|
| elt14lpo |
1:d0884279b41d
|
121
|
|
| elt14lpo |
1:d0884279b41d
|
122
|
/* Number of output samples is calculated */
|
| elt14lpo |
1:d0884279b41d
|
123
|
outBlockSize = (2u * srcALen) - 1u;
|
| elt14lpo |
1:d0884279b41d
|
124
|
|
| elt14lpo |
1:d0884279b41d
|
125
|
/* When srcALen > srcBLen, zero padding has to be done to srcB
|
| elt14lpo |
1:d0884279b41d
|
126
|
* to make their lengths equal.
|
| elt14lpo |
1:d0884279b41d
|
127
|
* Instead, (outBlockSize - (srcALen + srcBLen - 1))
|
| elt14lpo |
1:d0884279b41d
|
128
|
* number of output samples are made zero */
|
| elt14lpo |
1:d0884279b41d
|
129
|
j = outBlockSize - (srcALen + (srcBLen - 1u));
|
| elt14lpo |
1:d0884279b41d
|
130
|
|
| elt14lpo |
1:d0884279b41d
|
131
|
while(j > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
132
|
/* Zero is stored in the destination buffer */
|
| elt14lpo |
1:d0884279b41d
|
133
|
*pOut++ = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
134
|
|
| elt14lpo |
1:d0884279b41d
|
135
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
136
|
j--;
|
| elt14lpo |
1:d0884279b41d
|
137
|
}
|
| elt14lpo |
1:d0884279b41d
|
138
|
|
| elt14lpo |
1:d0884279b41d
|
139
|
} else {
|
| elt14lpo |
1:d0884279b41d
|
140
|
/* Initialization of inputA pointer */
|
| elt14lpo |
1:d0884279b41d
|
141
|
pIn1 = pSrcB;
|
| elt14lpo |
1:d0884279b41d
|
142
|
|
| elt14lpo |
1:d0884279b41d
|
143
|
/* Initialization of inputB pointer */
|
| elt14lpo |
1:d0884279b41d
|
144
|
pIn2 = pSrcA;
|
| elt14lpo |
1:d0884279b41d
|
145
|
|
| elt14lpo |
1:d0884279b41d
|
146
|
/* srcBLen is always considered as shorter or equal to srcALen */
|
| elt14lpo |
1:d0884279b41d
|
147
|
j = srcBLen;
|
| elt14lpo |
1:d0884279b41d
|
148
|
srcBLen = srcALen;
|
| elt14lpo |
1:d0884279b41d
|
149
|
srcALen = j;
|
| elt14lpo |
1:d0884279b41d
|
150
|
|
| elt14lpo |
1:d0884279b41d
|
151
|
/* CORR(x, y) = Reverse order(CORR(y, x)) */
|
| elt14lpo |
1:d0884279b41d
|
152
|
/* Hence set the destination pointer to point to the last output sample */
|
| elt14lpo |
1:d0884279b41d
|
153
|
pOut = pDst + ((srcALen + srcBLen) - 2u);
|
| elt14lpo |
1:d0884279b41d
|
154
|
|
| elt14lpo |
1:d0884279b41d
|
155
|
/* Destination address modifier is set to -1 */
|
| elt14lpo |
1:d0884279b41d
|
156
|
inc = -1;
|
| elt14lpo |
1:d0884279b41d
|
157
|
|
| elt14lpo |
1:d0884279b41d
|
158
|
}
|
| elt14lpo |
1:d0884279b41d
|
159
|
|
| elt14lpo |
1:d0884279b41d
|
160
|
/* The function is internally
|
| elt14lpo |
1:d0884279b41d
|
161
|
* divided into three parts according to the number of multiplications that has to be
|
| elt14lpo |
1:d0884279b41d
|
162
|
* taken place between inputA samples and inputB samples. In the first part of the
|
| elt14lpo |
1:d0884279b41d
|
163
|
* algorithm, the multiplications increase by one for every iteration.
|
| elt14lpo |
1:d0884279b41d
|
164
|
* In the second part of the algorithm, srcBLen number of multiplications are done.
|
| elt14lpo |
1:d0884279b41d
|
165
|
* In the third part of the algorithm, the multiplications decrease by one
|
| elt14lpo |
1:d0884279b41d
|
166
|
* for every iteration.*/
|
| elt14lpo |
1:d0884279b41d
|
167
|
/* The algorithm is implemented in three stages.
|
| elt14lpo |
1:d0884279b41d
|
168
|
* The loop counters of each stage is initiated here. */
|
| elt14lpo |
1:d0884279b41d
|
169
|
blockSize1 = srcBLen - 1u;
|
| elt14lpo |
1:d0884279b41d
|
170
|
blockSize2 = srcALen - (srcBLen - 1u);
|
| elt14lpo |
1:d0884279b41d
|
171
|
blockSize3 = blockSize1;
|
| elt14lpo |
1:d0884279b41d
|
172
|
|
| elt14lpo |
1:d0884279b41d
|
173
|
/* --------------------------
|
| elt14lpo |
1:d0884279b41d
|
174
|
* Initializations of stage1
|
| elt14lpo |
1:d0884279b41d
|
175
|
* -------------------------*/
|
| elt14lpo |
1:d0884279b41d
|
176
|
|
| elt14lpo |
1:d0884279b41d
|
177
|
/* sum = x[0] * y[srcBlen - 1]
|
| elt14lpo |
1:d0884279b41d
|
178
|
* sum = x[0] * y[srcBlen-2] + x[1] * y[srcBlen - 1]
|
| elt14lpo |
1:d0884279b41d
|
179
|
* ....
|
| elt14lpo |
1:d0884279b41d
|
180
|
* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1]
|
| elt14lpo |
1:d0884279b41d
|
181
|
*/
|
| elt14lpo |
1:d0884279b41d
|
182
|
|
| elt14lpo |
1:d0884279b41d
|
183
|
/* In this stage the MAC operations are increased by 1 for every iteration.
|
| elt14lpo |
1:d0884279b41d
|
184
|
The count variable holds the number of MAC operations performed */
|
| elt14lpo |
1:d0884279b41d
|
185
|
count = 1u;
|
| elt14lpo |
1:d0884279b41d
|
186
|
|
| elt14lpo |
1:d0884279b41d
|
187
|
/* Working pointer of inputA */
|
| elt14lpo |
1:d0884279b41d
|
188
|
px = pIn1;
|
| elt14lpo |
1:d0884279b41d
|
189
|
|
| elt14lpo |
1:d0884279b41d
|
190
|
/* Working pointer of inputB */
|
| elt14lpo |
1:d0884279b41d
|
191
|
pSrc1 = pIn2 + (srcBLen - 1u);
|
| elt14lpo |
1:d0884279b41d
|
192
|
py = pSrc1;
|
| elt14lpo |
1:d0884279b41d
|
193
|
|
| elt14lpo |
1:d0884279b41d
|
194
|
/* ------------------------
|
| elt14lpo |
1:d0884279b41d
|
195
|
* Stage1 process
|
| elt14lpo |
1:d0884279b41d
|
196
|
* ----------------------*/
|
| elt14lpo |
1:d0884279b41d
|
197
|
|
| elt14lpo |
1:d0884279b41d
|
198
|
/* The first stage starts here */
|
| elt14lpo |
1:d0884279b41d
|
199
|
while(blockSize1 > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
200
|
/* Accumulator is made zero for every iteration */
|
| elt14lpo |
1:d0884279b41d
|
201
|
sum = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
202
|
|
| elt14lpo |
1:d0884279b41d
|
203
|
/* Apply loop unrolling and compute 4 MACs simultaneously. */
|
| elt14lpo |
1:d0884279b41d
|
204
|
k = count >> 2u;
|
| elt14lpo |
1:d0884279b41d
|
205
|
|
| elt14lpo |
1:d0884279b41d
|
206
|
/* First part of the processing with loop unrolling. Compute 4 MACs at a time.
|
| elt14lpo |
1:d0884279b41d
|
207
|
** a second loop below computes MACs for the remaining 1 to 3 samples. */
|
| elt14lpo |
1:d0884279b41d
|
208
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
209
|
/* x[0] * y[srcBLen - 4] */
|
| elt14lpo |
1:d0884279b41d
|
210
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
211
|
/* x[1] * y[srcBLen - 3] */
|
| elt14lpo |
1:d0884279b41d
|
212
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
213
|
/* x[2] * y[srcBLen - 2] */
|
| elt14lpo |
1:d0884279b41d
|
214
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
215
|
/* x[3] * y[srcBLen - 1] */
|
| elt14lpo |
1:d0884279b41d
|
216
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
217
|
|
| elt14lpo |
1:d0884279b41d
|
218
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
219
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
220
|
}
|
| elt14lpo |
1:d0884279b41d
|
221
|
|
| elt14lpo |
1:d0884279b41d
|
222
|
/* If the count is not a multiple of 4, compute any remaining MACs here.
|
| elt14lpo |
1:d0884279b41d
|
223
|
** No loop unrolling is used. */
|
| elt14lpo |
1:d0884279b41d
|
224
|
k = count % 0x4u;
|
| elt14lpo |
1:d0884279b41d
|
225
|
|
| elt14lpo |
1:d0884279b41d
|
226
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
227
|
/* Perform the multiply-accumulate */
|
| elt14lpo |
1:d0884279b41d
|
228
|
/* x[0] * y[srcBLen - 1] */
|
| elt14lpo |
1:d0884279b41d
|
229
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
230
|
|
| elt14lpo |
1:d0884279b41d
|
231
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
232
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
233
|
}
|
| elt14lpo |
1:d0884279b41d
|
234
|
|
| elt14lpo |
1:d0884279b41d
|
235
|
/* Store the result in the accumulator in the destination buffer. */
|
| elt14lpo |
1:d0884279b41d
|
236
|
*pOut = sum;
|
| elt14lpo |
1:d0884279b41d
|
237
|
/* Destination pointer is updated according to the address modifier, inc */
|
| elt14lpo |
1:d0884279b41d
|
238
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
239
|
|
| elt14lpo |
1:d0884279b41d
|
240
|
/* Update the inputA and inputB pointers for next MAC calculation */
|
| elt14lpo |
1:d0884279b41d
|
241
|
py = pSrc1 - count;
|
| elt14lpo |
1:d0884279b41d
|
242
|
px = pIn1;
|
| elt14lpo |
1:d0884279b41d
|
243
|
|
| elt14lpo |
1:d0884279b41d
|
244
|
/* Increment the MAC count */
|
| elt14lpo |
1:d0884279b41d
|
245
|
count++;
|
| elt14lpo |
1:d0884279b41d
|
246
|
|
| elt14lpo |
1:d0884279b41d
|
247
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
248
|
blockSize1--;
|
| elt14lpo |
1:d0884279b41d
|
249
|
}
|
| elt14lpo |
1:d0884279b41d
|
250
|
|
| elt14lpo |
1:d0884279b41d
|
251
|
/* --------------------------
|
| elt14lpo |
1:d0884279b41d
|
252
|
* Initializations of stage2
|
| elt14lpo |
1:d0884279b41d
|
253
|
* ------------------------*/
|
| elt14lpo |
1:d0884279b41d
|
254
|
|
| elt14lpo |
1:d0884279b41d
|
255
|
/* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1]
|
| elt14lpo |
1:d0884279b41d
|
256
|
* sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1]
|
| elt14lpo |
1:d0884279b41d
|
257
|
* ....
|
| elt14lpo |
1:d0884279b41d
|
258
|
* sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
|
| elt14lpo |
1:d0884279b41d
|
259
|
*/
|
| elt14lpo |
1:d0884279b41d
|
260
|
|
| elt14lpo |
1:d0884279b41d
|
261
|
/* Working pointer of inputA */
|
| elt14lpo |
1:d0884279b41d
|
262
|
px = pIn1;
|
| elt14lpo |
1:d0884279b41d
|
263
|
|
| elt14lpo |
1:d0884279b41d
|
264
|
/* Working pointer of inputB */
|
| elt14lpo |
1:d0884279b41d
|
265
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
266
|
|
| elt14lpo |
1:d0884279b41d
|
267
|
/* count is index by which the pointer pIn1 to be incremented */
|
| elt14lpo |
1:d0884279b41d
|
268
|
count = 1u;
|
| elt14lpo |
1:d0884279b41d
|
269
|
|
| elt14lpo |
1:d0884279b41d
|
270
|
/* -------------------
|
| elt14lpo |
1:d0884279b41d
|
271
|
* Stage2 process
|
| elt14lpo |
1:d0884279b41d
|
272
|
* ------------------*/
|
| elt14lpo |
1:d0884279b41d
|
273
|
|
| elt14lpo |
1:d0884279b41d
|
274
|
/* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
|
| elt14lpo |
1:d0884279b41d
|
275
|
* So, to loop unroll over blockSize2,
|
| elt14lpo |
1:d0884279b41d
|
276
|
* srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */
|
| elt14lpo |
1:d0884279b41d
|
277
|
if(srcBLen >= 4u) {
|
| elt14lpo |
1:d0884279b41d
|
278
|
/* Loop unroll over blockSize2, by 4 */
|
| elt14lpo |
1:d0884279b41d
|
279
|
blkCnt = blockSize2 >> 2u;
|
| elt14lpo |
1:d0884279b41d
|
280
|
|
| elt14lpo |
1:d0884279b41d
|
281
|
while(blkCnt > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
282
|
/* Set all accumulators to zero */
|
| elt14lpo |
1:d0884279b41d
|
283
|
acc0 = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
284
|
acc1 = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
285
|
acc2 = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
286
|
acc3 = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
287
|
|
| elt14lpo |
1:d0884279b41d
|
288
|
/* read x[0], x[1], x[2] samples */
|
| elt14lpo |
1:d0884279b41d
|
289
|
x0 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
290
|
x1 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
291
|
x2 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
292
|
|
| elt14lpo |
1:d0884279b41d
|
293
|
/* Apply loop unrolling and compute 4 MACs simultaneously. */
|
| elt14lpo |
1:d0884279b41d
|
294
|
k = srcBLen >> 2u;
|
| elt14lpo |
1:d0884279b41d
|
295
|
|
| elt14lpo |
1:d0884279b41d
|
296
|
/* First part of the processing with loop unrolling. Compute 4 MACs at a time.
|
| elt14lpo |
1:d0884279b41d
|
297
|
** a second loop below computes MACs for the remaining 1 to 3 samples. */
|
| elt14lpo |
1:d0884279b41d
|
298
|
do {
|
| elt14lpo |
1:d0884279b41d
|
299
|
/* Read y[0] sample */
|
| elt14lpo |
1:d0884279b41d
|
300
|
c0 = *(py++);
|
| elt14lpo |
1:d0884279b41d
|
301
|
|
| elt14lpo |
1:d0884279b41d
|
302
|
/* Read x[3] sample */
|
| elt14lpo |
1:d0884279b41d
|
303
|
x3 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
304
|
|
| elt14lpo |
1:d0884279b41d
|
305
|
/* Perform the multiply-accumulate */
|
| elt14lpo |
1:d0884279b41d
|
306
|
/* acc0 += x[0] * y[0] */
|
| elt14lpo |
1:d0884279b41d
|
307
|
acc0 += x0 * c0;
|
| elt14lpo |
1:d0884279b41d
|
308
|
/* acc1 += x[1] * y[0] */
|
| elt14lpo |
1:d0884279b41d
|
309
|
acc1 += x1 * c0;
|
| elt14lpo |
1:d0884279b41d
|
310
|
/* acc2 += x[2] * y[0] */
|
| elt14lpo |
1:d0884279b41d
|
311
|
acc2 += x2 * c0;
|
| elt14lpo |
1:d0884279b41d
|
312
|
/* acc3 += x[3] * y[0] */
|
| elt14lpo |
1:d0884279b41d
|
313
|
acc3 += x3 * c0;
|
| elt14lpo |
1:d0884279b41d
|
314
|
|
| elt14lpo |
1:d0884279b41d
|
315
|
/* Read y[1] sample */
|
| elt14lpo |
1:d0884279b41d
|
316
|
c0 = *(py++);
|
| elt14lpo |
1:d0884279b41d
|
317
|
|
| elt14lpo |
1:d0884279b41d
|
318
|
/* Read x[4] sample */
|
| elt14lpo |
1:d0884279b41d
|
319
|
x0 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
320
|
|
| elt14lpo |
1:d0884279b41d
|
321
|
/* Perform the multiply-accumulate */
|
| elt14lpo |
1:d0884279b41d
|
322
|
/* acc0 += x[1] * y[1] */
|
| elt14lpo |
1:d0884279b41d
|
323
|
acc0 += x1 * c0;
|
| elt14lpo |
1:d0884279b41d
|
324
|
/* acc1 += x[2] * y[1] */
|
| elt14lpo |
1:d0884279b41d
|
325
|
acc1 += x2 * c0;
|
| elt14lpo |
1:d0884279b41d
|
326
|
/* acc2 += x[3] * y[1] */
|
| elt14lpo |
1:d0884279b41d
|
327
|
acc2 += x3 * c0;
|
| elt14lpo |
1:d0884279b41d
|
328
|
/* acc3 += x[4] * y[1] */
|
| elt14lpo |
1:d0884279b41d
|
329
|
acc3 += x0 * c0;
|
| elt14lpo |
1:d0884279b41d
|
330
|
|
| elt14lpo |
1:d0884279b41d
|
331
|
/* Read y[2] sample */
|
| elt14lpo |
1:d0884279b41d
|
332
|
c0 = *(py++);
|
| elt14lpo |
1:d0884279b41d
|
333
|
|
| elt14lpo |
1:d0884279b41d
|
334
|
/* Read x[5] sample */
|
| elt14lpo |
1:d0884279b41d
|
335
|
x1 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
336
|
|
| elt14lpo |
1:d0884279b41d
|
337
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
338
|
/* acc0 += x[2] * y[2] */
|
| elt14lpo |
1:d0884279b41d
|
339
|
acc0 += x2 * c0;
|
| elt14lpo |
1:d0884279b41d
|
340
|
/* acc1 += x[3] * y[2] */
|
| elt14lpo |
1:d0884279b41d
|
341
|
acc1 += x3 * c0;
|
| elt14lpo |
1:d0884279b41d
|
342
|
/* acc2 += x[4] * y[2] */
|
| elt14lpo |
1:d0884279b41d
|
343
|
acc2 += x0 * c0;
|
| elt14lpo |
1:d0884279b41d
|
344
|
/* acc3 += x[5] * y[2] */
|
| elt14lpo |
1:d0884279b41d
|
345
|
acc3 += x1 * c0;
|
| elt14lpo |
1:d0884279b41d
|
346
|
|
| elt14lpo |
1:d0884279b41d
|
347
|
/* Read y[3] sample */
|
| elt14lpo |
1:d0884279b41d
|
348
|
c0 = *(py++);
|
| elt14lpo |
1:d0884279b41d
|
349
|
|
| elt14lpo |
1:d0884279b41d
|
350
|
/* Read x[6] sample */
|
| elt14lpo |
1:d0884279b41d
|
351
|
x2 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
352
|
|
| elt14lpo |
1:d0884279b41d
|
353
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
354
|
/* acc0 += x[3] * y[3] */
|
| elt14lpo |
1:d0884279b41d
|
355
|
acc0 += x3 * c0;
|
| elt14lpo |
1:d0884279b41d
|
356
|
/* acc1 += x[4] * y[3] */
|
| elt14lpo |
1:d0884279b41d
|
357
|
acc1 += x0 * c0;
|
| elt14lpo |
1:d0884279b41d
|
358
|
/* acc2 += x[5] * y[3] */
|
| elt14lpo |
1:d0884279b41d
|
359
|
acc2 += x1 * c0;
|
| elt14lpo |
1:d0884279b41d
|
360
|
/* acc3 += x[6] * y[3] */
|
| elt14lpo |
1:d0884279b41d
|
361
|
acc3 += x2 * c0;
|
| elt14lpo |
1:d0884279b41d
|
362
|
|
| elt14lpo |
1:d0884279b41d
|
363
|
|
| elt14lpo |
1:d0884279b41d
|
364
|
} while(--k);
|
| elt14lpo |
1:d0884279b41d
|
365
|
|
| elt14lpo |
1:d0884279b41d
|
366
|
/* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
|
| elt14lpo |
1:d0884279b41d
|
367
|
** No loop unrolling is used. */
|
| elt14lpo |
1:d0884279b41d
|
368
|
k = srcBLen % 0x4u;
|
| elt14lpo |
1:d0884279b41d
|
369
|
|
| elt14lpo |
1:d0884279b41d
|
370
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
371
|
/* Read y[4] sample */
|
| elt14lpo |
1:d0884279b41d
|
372
|
c0 = *(py++);
|
| elt14lpo |
1:d0884279b41d
|
373
|
|
| elt14lpo |
1:d0884279b41d
|
374
|
/* Read x[7] sample */
|
| elt14lpo |
1:d0884279b41d
|
375
|
x3 = *(px++);
|
| elt14lpo |
1:d0884279b41d
|
376
|
|
| elt14lpo |
1:d0884279b41d
|
377
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
378
|
/* acc0 += x[4] * y[4] */
|
| elt14lpo |
1:d0884279b41d
|
379
|
acc0 += x0 * c0;
|
| elt14lpo |
1:d0884279b41d
|
380
|
/* acc1 += x[5] * y[4] */
|
| elt14lpo |
1:d0884279b41d
|
381
|
acc1 += x1 * c0;
|
| elt14lpo |
1:d0884279b41d
|
382
|
/* acc2 += x[6] * y[4] */
|
| elt14lpo |
1:d0884279b41d
|
383
|
acc2 += x2 * c0;
|
| elt14lpo |
1:d0884279b41d
|
384
|
/* acc3 += x[7] * y[4] */
|
| elt14lpo |
1:d0884279b41d
|
385
|
acc3 += x3 * c0;
|
| elt14lpo |
1:d0884279b41d
|
386
|
|
| elt14lpo |
1:d0884279b41d
|
387
|
/* Reuse the present samples for the next MAC */
|
| elt14lpo |
1:d0884279b41d
|
388
|
x0 = x1;
|
| elt14lpo |
1:d0884279b41d
|
389
|
x1 = x2;
|
| elt14lpo |
1:d0884279b41d
|
390
|
x2 = x3;
|
| elt14lpo |
1:d0884279b41d
|
391
|
|
| elt14lpo |
1:d0884279b41d
|
392
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
393
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
394
|
}
|
| elt14lpo |
1:d0884279b41d
|
395
|
|
| elt14lpo |
1:d0884279b41d
|
396
|
/* Store the result in the accumulator in the destination buffer. */
|
| elt14lpo |
1:d0884279b41d
|
397
|
*pOut = acc0;
|
| elt14lpo |
1:d0884279b41d
|
398
|
/* Destination pointer is updated according to the address modifier, inc */
|
| elt14lpo |
1:d0884279b41d
|
399
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
400
|
|
| elt14lpo |
1:d0884279b41d
|
401
|
*pOut = acc1;
|
| elt14lpo |
1:d0884279b41d
|
402
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
403
|
|
| elt14lpo |
1:d0884279b41d
|
404
|
*pOut = acc2;
|
| elt14lpo |
1:d0884279b41d
|
405
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
406
|
|
| elt14lpo |
1:d0884279b41d
|
407
|
*pOut = acc3;
|
| elt14lpo |
1:d0884279b41d
|
408
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
409
|
|
| elt14lpo |
1:d0884279b41d
|
410
|
/* Update the inputA and inputB pointers for next MAC calculation */
|
| elt14lpo |
1:d0884279b41d
|
411
|
px = pIn1 + (count * 4u);
|
| elt14lpo |
1:d0884279b41d
|
412
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
413
|
|
| elt14lpo |
1:d0884279b41d
|
414
|
/* Increment the pointer pIn1 index, count by 1 */
|
| elt14lpo |
1:d0884279b41d
|
415
|
count++;
|
| elt14lpo |
1:d0884279b41d
|
416
|
|
| elt14lpo |
1:d0884279b41d
|
417
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
418
|
blkCnt--;
|
| elt14lpo |
1:d0884279b41d
|
419
|
}
|
| elt14lpo |
1:d0884279b41d
|
420
|
|
| elt14lpo |
1:d0884279b41d
|
421
|
/* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.
|
| elt14lpo |
1:d0884279b41d
|
422
|
** No loop unrolling is used. */
|
| elt14lpo |
1:d0884279b41d
|
423
|
blkCnt = blockSize2 % 0x4u;
|
| elt14lpo |
1:d0884279b41d
|
424
|
|
| elt14lpo |
1:d0884279b41d
|
425
|
while(blkCnt > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
426
|
/* Accumulator is made zero for every iteration */
|
| elt14lpo |
1:d0884279b41d
|
427
|
sum = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
428
|
|
| elt14lpo |
1:d0884279b41d
|
429
|
/* Apply loop unrolling and compute 4 MACs simultaneously. */
|
| elt14lpo |
1:d0884279b41d
|
430
|
k = srcBLen >> 2u;
|
| elt14lpo |
1:d0884279b41d
|
431
|
|
| elt14lpo |
1:d0884279b41d
|
432
|
/* First part of the processing with loop unrolling. Compute 4 MACs at a time.
|
| elt14lpo |
1:d0884279b41d
|
433
|
** a second loop below computes MACs for the remaining 1 to 3 samples. */
|
| elt14lpo |
1:d0884279b41d
|
434
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
435
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
436
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
437
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
438
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
439
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
440
|
|
| elt14lpo |
1:d0884279b41d
|
441
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
442
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
443
|
}
|
| elt14lpo |
1:d0884279b41d
|
444
|
|
| elt14lpo |
1:d0884279b41d
|
445
|
/* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
|
| elt14lpo |
1:d0884279b41d
|
446
|
** No loop unrolling is used. */
|
| elt14lpo |
1:d0884279b41d
|
447
|
k = srcBLen % 0x4u;
|
| elt14lpo |
1:d0884279b41d
|
448
|
|
| elt14lpo |
1:d0884279b41d
|
449
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
450
|
/* Perform the multiply-accumulate */
|
| elt14lpo |
1:d0884279b41d
|
451
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
452
|
|
| elt14lpo |
1:d0884279b41d
|
453
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
454
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
455
|
}
|
| elt14lpo |
1:d0884279b41d
|
456
|
|
| elt14lpo |
1:d0884279b41d
|
457
|
/* Store the result in the accumulator in the destination buffer. */
|
| elt14lpo |
1:d0884279b41d
|
458
|
*pOut = sum;
|
| elt14lpo |
1:d0884279b41d
|
459
|
/* Destination pointer is updated according to the address modifier, inc */
|
| elt14lpo |
1:d0884279b41d
|
460
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
461
|
|
| elt14lpo |
1:d0884279b41d
|
462
|
/* Update the inputA and inputB pointers for next MAC calculation */
|
| elt14lpo |
1:d0884279b41d
|
463
|
px = pIn1 + count;
|
| elt14lpo |
1:d0884279b41d
|
464
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
465
|
|
| elt14lpo |
1:d0884279b41d
|
466
|
/* Increment the pointer pIn1 index, count by 1 */
|
| elt14lpo |
1:d0884279b41d
|
467
|
count++;
|
| elt14lpo |
1:d0884279b41d
|
468
|
|
| elt14lpo |
1:d0884279b41d
|
469
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
470
|
blkCnt--;
|
| elt14lpo |
1:d0884279b41d
|
471
|
}
|
| elt14lpo |
1:d0884279b41d
|
472
|
} else {
|
| elt14lpo |
1:d0884279b41d
|
473
|
/* If the srcBLen is not a multiple of 4,
|
| elt14lpo |
1:d0884279b41d
|
474
|
* the blockSize2 loop cannot be unrolled by 4 */
|
| elt14lpo |
1:d0884279b41d
|
475
|
blkCnt = blockSize2;
|
| elt14lpo |
1:d0884279b41d
|
476
|
|
| elt14lpo |
1:d0884279b41d
|
477
|
while(blkCnt > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
478
|
/* Accumulator is made zero for every iteration */
|
| elt14lpo |
1:d0884279b41d
|
479
|
sum = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
480
|
|
| elt14lpo |
1:d0884279b41d
|
481
|
/* Loop over srcBLen */
|
| elt14lpo |
1:d0884279b41d
|
482
|
k = srcBLen;
|
| elt14lpo |
1:d0884279b41d
|
483
|
|
| elt14lpo |
1:d0884279b41d
|
484
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
485
|
/* Perform the multiply-accumulate */
|
| elt14lpo |
1:d0884279b41d
|
486
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
487
|
|
| elt14lpo |
1:d0884279b41d
|
488
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
489
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
490
|
}
|
| elt14lpo |
1:d0884279b41d
|
491
|
|
| elt14lpo |
1:d0884279b41d
|
492
|
/* Store the result in the accumulator in the destination buffer. */
|
| elt14lpo |
1:d0884279b41d
|
493
|
*pOut = sum;
|
| elt14lpo |
1:d0884279b41d
|
494
|
/* Destination pointer is updated according to the address modifier, inc */
|
| elt14lpo |
1:d0884279b41d
|
495
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
496
|
|
| elt14lpo |
1:d0884279b41d
|
497
|
/* Update the inputA and inputB pointers for next MAC calculation */
|
| elt14lpo |
1:d0884279b41d
|
498
|
px = pIn1 + count;
|
| elt14lpo |
1:d0884279b41d
|
499
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
500
|
|
| elt14lpo |
1:d0884279b41d
|
501
|
/* Increment the pointer pIn1 index, count by 1 */
|
| elt14lpo |
1:d0884279b41d
|
502
|
count++;
|
| elt14lpo |
1:d0884279b41d
|
503
|
|
| elt14lpo |
1:d0884279b41d
|
504
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
505
|
blkCnt--;
|
| elt14lpo |
1:d0884279b41d
|
506
|
}
|
| elt14lpo |
1:d0884279b41d
|
507
|
}
|
| elt14lpo |
1:d0884279b41d
|
508
|
|
| elt14lpo |
1:d0884279b41d
|
509
|
/* --------------------------
|
| elt14lpo |
1:d0884279b41d
|
510
|
* Initializations of stage3
|
| elt14lpo |
1:d0884279b41d
|
511
|
* -------------------------*/
|
| elt14lpo |
1:d0884279b41d
|
512
|
|
| elt14lpo |
1:d0884279b41d
|
513
|
/* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
|
| elt14lpo |
1:d0884279b41d
|
514
|
* sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]
|
| elt14lpo |
1:d0884279b41d
|
515
|
* ....
|
| elt14lpo |
1:d0884279b41d
|
516
|
* sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1]
|
| elt14lpo |
1:d0884279b41d
|
517
|
* sum += x[srcALen-1] * y[0]
|
| elt14lpo |
1:d0884279b41d
|
518
|
*/
|
| elt14lpo |
1:d0884279b41d
|
519
|
|
| elt14lpo |
1:d0884279b41d
|
520
|
/* In this stage the MAC operations are decreased by 1 for every iteration.
|
| elt14lpo |
1:d0884279b41d
|
521
|
The count variable holds the number of MAC operations performed */
|
| elt14lpo |
1:d0884279b41d
|
522
|
count = srcBLen - 1u;
|
| elt14lpo |
1:d0884279b41d
|
523
|
|
| elt14lpo |
1:d0884279b41d
|
524
|
/* Working pointer of inputA */
|
| elt14lpo |
1:d0884279b41d
|
525
|
pSrc1 = pIn1 + (srcALen - (srcBLen - 1u));
|
| elt14lpo |
1:d0884279b41d
|
526
|
px = pSrc1;
|
| elt14lpo |
1:d0884279b41d
|
527
|
|
| elt14lpo |
1:d0884279b41d
|
528
|
/* Working pointer of inputB */
|
| elt14lpo |
1:d0884279b41d
|
529
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
530
|
|
| elt14lpo |
1:d0884279b41d
|
531
|
/* -------------------
|
| elt14lpo |
1:d0884279b41d
|
532
|
* Stage3 process
|
| elt14lpo |
1:d0884279b41d
|
533
|
* ------------------*/
|
| elt14lpo |
1:d0884279b41d
|
534
|
|
| elt14lpo |
1:d0884279b41d
|
535
|
while(blockSize3 > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
536
|
/* Accumulator is made zero for every iteration */
|
| elt14lpo |
1:d0884279b41d
|
537
|
sum = 0.0f;
|
| elt14lpo |
1:d0884279b41d
|
538
|
|
| elt14lpo |
1:d0884279b41d
|
539
|
/* Apply loop unrolling and compute 4 MACs simultaneously. */
|
| elt14lpo |
1:d0884279b41d
|
540
|
k = count >> 2u;
|
| elt14lpo |
1:d0884279b41d
|
541
|
|
| elt14lpo |
1:d0884279b41d
|
542
|
/* First part of the processing with loop unrolling. Compute 4 MACs at a time.
|
| elt14lpo |
1:d0884279b41d
|
543
|
** a second loop below computes MACs for the remaining 1 to 3 samples. */
|
| elt14lpo |
1:d0884279b41d
|
544
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
545
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
546
|
/* sum += x[srcALen - srcBLen + 4] * y[3] */
|
| elt14lpo |
1:d0884279b41d
|
547
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
548
|
/* sum += x[srcALen - srcBLen + 3] * y[2] */
|
| elt14lpo |
1:d0884279b41d
|
549
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
550
|
/* sum += x[srcALen - srcBLen + 2] * y[1] */
|
| elt14lpo |
1:d0884279b41d
|
551
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
552
|
/* sum += x[srcALen - srcBLen + 1] * y[0] */
|
| elt14lpo |
1:d0884279b41d
|
553
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
554
|
|
| elt14lpo |
1:d0884279b41d
|
555
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
556
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
557
|
}
|
| elt14lpo |
1:d0884279b41d
|
558
|
|
| elt14lpo |
1:d0884279b41d
|
559
|
/* If the count is not a multiple of 4, compute any remaining MACs here.
|
| elt14lpo |
1:d0884279b41d
|
560
|
** No loop unrolling is used. */
|
| elt14lpo |
1:d0884279b41d
|
561
|
k = count % 0x4u;
|
| elt14lpo |
1:d0884279b41d
|
562
|
|
| elt14lpo |
1:d0884279b41d
|
563
|
while(k > 0u) {
|
| elt14lpo |
1:d0884279b41d
|
564
|
/* Perform the multiply-accumulates */
|
| elt14lpo |
1:d0884279b41d
|
565
|
sum += *px++ * *py++;
|
| elt14lpo |
1:d0884279b41d
|
566
|
|
| elt14lpo |
1:d0884279b41d
|
567
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
568
|
k--;
|
| elt14lpo |
1:d0884279b41d
|
569
|
}
|
| elt14lpo |
1:d0884279b41d
|
570
|
|
| elt14lpo |
1:d0884279b41d
|
571
|
/* Store the result in the accumulator in the destination buffer. */
|
| elt14lpo |
1:d0884279b41d
|
572
|
*pOut = sum;
|
| elt14lpo |
1:d0884279b41d
|
573
|
/* Destination pointer is updated according to the address modifier, inc */
|
| elt14lpo |
1:d0884279b41d
|
574
|
pOut += inc;
|
| elt14lpo |
1:d0884279b41d
|
575
|
|
| elt14lpo |
1:d0884279b41d
|
576
|
/* Update the inputA and inputB pointers for next MAC calculation */
|
| elt14lpo |
1:d0884279b41d
|
577
|
px = ++pSrc1;
|
| elt14lpo |
1:d0884279b41d
|
578
|
py = pIn2;
|
| elt14lpo |
1:d0884279b41d
|
579
|
|
| elt14lpo |
1:d0884279b41d
|
580
|
/* Decrement the MAC count */
|
| elt14lpo |
1:d0884279b41d
|
581
|
count--;
|
| elt14lpo |
1:d0884279b41d
|
582
|
|
| elt14lpo |
1:d0884279b41d
|
583
|
/* Decrement the loop counter */
|
| elt14lpo |
1:d0884279b41d
|
584
|
blockSize3--;
|
| elt14lpo |
1:d0884279b41d
|
585
|
}
|
| elt14lpo |
1:d0884279b41d
|
586
|
|
| elt14lpo |
1:d0884279b41d
|
587
|
}
|
| elt14lpo |
1:d0884279b41d
|
588
|
|
| elt14lpo |
1:d0884279b41d
|
589
|
/**
|
| elt14lpo |
1:d0884279b41d
|
590
|
* @} end of Corr group
|
| elt14lpo |
1:d0884279b41d
|
591
|
*/
|
Elektronikprojekt Grupp 13