1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
2 <html xmlns="http://www.w3.org/1999/xhtml">
4 <meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
5 <title>CMSIS DSP Software Library: arm_correlate_q31.c Source File</title>
6 <link href="tabs.css" rel="stylesheet" type="text/css"/>
7 <link href="search/search.css" rel="stylesheet" type="text/css"/>
8 <script type="text/javaScript" src="search/search.js"></script>
9 <link href="doxygen.css" rel="stylesheet" type="text/css"/>
11 <body onload='searchBox.OnSelectItem(0);'>
12 <!-- Generated by Doxygen 1.7.2 -->
13 <script type="text/javascript"><!--
14 var searchBox = new SearchBox("searchBox", "search",false,'Search');
16 <div class="navigation" id="top">
19 <li><a href="index.html"><span>Main Page</span></a></li>
20 <li><a href="modules.html"><span>Modules</span></a></li>
21 <li><a href="annotated.html"><span>Data Structures</span></a></li>
22 <li class="current"><a href="files.html"><span>Files</span></a></li>
23 <li><a href="examples.html"><span>Examples</span></a></li>
25 <div id="MSearchBox" class="MSearchBoxInactive">
27 <img id="MSearchSelect" src="search/mag_sel.png"
28 onmouseover="return searchBox.OnSearchSelectShow()"
29 onmouseout="return searchBox.OnSearchSelectHide()"
31 <input type="text" id="MSearchField" value="Search" accesskey="S"
32 onfocus="searchBox.OnSearchFieldFocus(true)"
33 onblur="searchBox.OnSearchFieldFocus(false)"
34 onkeyup="searchBox.OnSearchFieldChange(event)"/>
35 </span><span class="right">
36 <a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
44 <li><a href="files.html"><span>File List</span></a></li>
45 <li><a href="globals.html"><span>Globals</span></a></li>
49 <div class="headertitle">
50 <h1>arm_correlate_q31.c</h1> </div>
52 <div class="contents">
53 <a href="arm__correlate__q31_8c.html">Go to the documentation of this file.</a><div class="fragment"><pre class="fragment"><a name="l00001"></a>00001 <span class="comment">/* ---------------------------------------------------------------------- </span>
54 <a name="l00002"></a>00002 <span class="comment">* Copyright (C) 2010 ARM Limited. All rights reserved. </span>
55 <a name="l00003"></a>00003 <span class="comment">* </span>
56 <a name="l00004"></a>00004 <span class="comment">* $Date: 15. July 2011 </span>
57 <a name="l00005"></a>00005 <span class="comment">* $Revision: V1.0.10 </span>
58 <a name="l00006"></a>00006 <span class="comment">* </span>
59 <a name="l00007"></a>00007 <span class="comment">* Project: CMSIS DSP Library </span>
60 <a name="l00008"></a>00008 <span class="comment">* Title: arm_correlate_q31.c </span>
61 <a name="l00009"></a>00009 <span class="comment">* </span>
62 <a name="l00010"></a>00010 <span class="comment">* Description: Correlation of Q31 sequences. </span>
63 <a name="l00011"></a>00011 <span class="comment">* </span>
64 <a name="l00012"></a>00012 <span class="comment">* Target Processor: Cortex-M4/Cortex-M3/Cortex-M0</span>
65 <a name="l00013"></a>00013 <span class="comment">* </span>
66 <a name="l00014"></a>00014 <span class="comment">* Version 1.0.10 2011/7/15 </span>
67 <a name="l00015"></a>00015 <span class="comment">* Big Endian support added and Merged M0 and M3/M4 Source code. </span>
68 <a name="l00016"></a>00016 <span class="comment">* </span>
69 <a name="l00017"></a>00017 <span class="comment">* Version 1.0.3 2010/11/29 </span>
70 <a name="l00018"></a>00018 <span class="comment">* Re-organized the CMSIS folders and updated documentation. </span>
71 <a name="l00019"></a>00019 <span class="comment">* </span>
72 <a name="l00020"></a>00020 <span class="comment">* Version 1.0.2 2010/11/11 </span>
73 <a name="l00021"></a>00021 <span class="comment">* Documentation updated. </span>
74 <a name="l00022"></a>00022 <span class="comment">* </span>
75 <a name="l00023"></a>00023 <span class="comment">* Version 1.0.1 2010/10/05 </span>
76 <a name="l00024"></a>00024 <span class="comment">* Production release and review comments incorporated. </span>
77 <a name="l00025"></a>00025 <span class="comment">* </span>
78 <a name="l00026"></a>00026 <span class="comment">* Version 1.0.0 2010/09/20 </span>
79 <a name="l00027"></a>00027 <span class="comment">* Production release and review comments incorporated </span>
80 <a name="l00028"></a>00028 <span class="comment">* </span>
81 <a name="l00029"></a>00029 <span class="comment">* Version 0.0.7 2010/06/10 </span>
82 <a name="l00030"></a>00030 <span class="comment">* Misra-C changes done </span>
83 <a name="l00031"></a>00031 <span class="comment">* </span>
84 <a name="l00032"></a>00032 <span class="comment">* -------------------------------------------------------------------- */</span>
85 <a name="l00033"></a>00033
86 <a name="l00034"></a>00034 <span class="preprocessor">#include "<a class="code" href="arm__math_8h.html">arm_math.h</a>"</span>
87 <a name="l00035"></a>00035
88 <a name="l00071"></a><a class="code" href="group___corr.html#ga1367dc6c80476406c951e68d7fac4e8c">00071</a> <span class="keywordtype">void</span> <a class="code" href="group___corr.html#ga1367dc6c80476406c951e68d7fac4e8c" title="Correlation of Q31 sequences.">arm_correlate_q31</a>(
89 <a name="l00072"></a>00072 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> * pSrcA,
90 <a name="l00073"></a>00073 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>,
91 <a name="l00074"></a>00074 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> * pSrcB,
92 <a name="l00075"></a>00075 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>,
93 <a name="l00076"></a>00076 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> * pDst)
94 <a name="l00077"></a>00077 {
95 <a name="l00078"></a>00078
96 <a name="l00079"></a>00079 <span class="preprocessor">#ifndef ARM_MATH_CM0</span>
97 <a name="l00080"></a>00080 <span class="preprocessor"></span>
98 <a name="l00081"></a>00081 <span class="comment">/* Run the below code for Cortex-M4 and Cortex-M3 */</span>
99 <a name="l00082"></a>00082
100 <a name="l00083"></a>00083 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pIn1; <span class="comment">/* inputA pointer */</span>
101 <a name="l00084"></a>00084 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pIn2; <span class="comment">/* inputB pointer */</span>
102 <a name="l00085"></a>00085 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pOut = pDst; <span class="comment">/* output pointer */</span>
103 <a name="l00086"></a>00086 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *px; <span class="comment">/* Intermediate inputA pointer */</span>
104 <a name="l00087"></a>00087 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *py; <span class="comment">/* Intermediate inputB pointer */</span>
105 <a name="l00088"></a>00088 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pSrc1; <span class="comment">/* Intermediate pointers */</span>
106 <a name="l00089"></a>00089 <a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a> sum, acc0, acc1, acc2, acc3; <span class="comment">/* Accumulators */</span>
107 <a name="l00090"></a>00090 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> x0, x1, x2, x3, c0; <span class="comment">/* temporary variables for holding input and coefficient values */</span>
108 <a name="l00091"></a>00091 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; <span class="comment">/* loop counter */</span>
109 <a name="l00092"></a>00092 int32_t inc = 1; <span class="comment">/* Destination address modifier */</span>
110 <a name="l00093"></a>00093
111 <a name="l00094"></a>00094
112 <a name="l00095"></a>00095 <span class="comment">/* The algorithm implementation is based on the lengths of the inputs. */</span>
113 <a name="l00096"></a>00096 <span class="comment">/* srcB is always made to slide across srcA. */</span>
114 <a name="l00097"></a>00097 <span class="comment">/* So srcBLen is always considered as shorter or equal to srcALen */</span>
115 <a name="l00098"></a>00098 <span class="comment">/* But CORR(x, y) is reverse of CORR(y, x) */</span>
116 <a name="l00099"></a>00099 <span class="comment">/* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */</span>
117 <a name="l00100"></a>00100 <span class="comment">/* and the destination pointer modifier, inc is set to -1 */</span>
118 <a name="l00101"></a>00101 <span class="comment">/* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */</span>
119 <a name="l00102"></a>00102 <span class="comment">/* But to improve the performance, </span>
120 <a name="l00103"></a>00103 <span class="comment"> * we include zeroes in the output instead of zero padding either of the the inputs*/</span>
121 <a name="l00104"></a>00104 <span class="comment">/* If srcALen > srcBLen, </span>
122 <a name="l00105"></a>00105 <span class="comment"> * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */</span>
123 <a name="l00106"></a>00106 <span class="comment">/* If srcALen < srcBLen, </span>
124 <a name="l00107"></a>00107 <span class="comment"> * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */</span>
125 <a name="l00108"></a>00108 <span class="keywordflow">if</span>(srcALen >= srcBLen)
126 <a name="l00109"></a>00109 {
127 <a name="l00110"></a>00110 <span class="comment">/* Initialization of inputA pointer */</span>
128 <a name="l00111"></a>00111 pIn1 = (pSrcA);
129 <a name="l00112"></a>00112
130 <a name="l00113"></a>00113 <span class="comment">/* Initialization of inputB pointer */</span>
131 <a name="l00114"></a>00114 pIn2 = (pSrcB);
132 <a name="l00115"></a>00115
133 <a name="l00116"></a>00116 <span class="comment">/* Number of output samples is calculated */</span>
134 <a name="l00117"></a>00117 outBlockSize = (2u * <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>) - 1u;
135 <a name="l00118"></a>00118
136 <a name="l00119"></a>00119 <span class="comment">/* When srcALen > srcBLen, zero padding is done to srcB </span>
137 <a name="l00120"></a>00120 <span class="comment"> * to make their lengths equal. </span>
138 <a name="l00121"></a>00121 <span class="comment"> * Instead, (outBlockSize - (srcALen + srcBLen - 1)) </span>
139 <a name="l00122"></a>00122 <span class="comment"> * number of output samples are made zero */</span>
140 <a name="l00123"></a>00123 j = outBlockSize - (srcALen + (srcBLen - 1u));
141 <a name="l00124"></a>00124
142 <a name="l00125"></a>00125 <span class="comment">/* Updating the pointer position to non zero value */</span>
143 <a name="l00126"></a>00126 pOut += j;
144 <a name="l00127"></a>00127
145 <a name="l00128"></a>00128 }
146 <a name="l00129"></a>00129 <span class="keywordflow">else</span>
147 <a name="l00130"></a>00130 {
148 <a name="l00131"></a>00131 <span class="comment">/* Initialization of inputA pointer */</span>
149 <a name="l00132"></a>00132 pIn1 = (pSrcB);
150 <a name="l00133"></a>00133
151 <a name="l00134"></a>00134 <span class="comment">/* Initialization of inputB pointer */</span>
152 <a name="l00135"></a>00135 pIn2 = (pSrcA);
153 <a name="l00136"></a>00136
154 <a name="l00137"></a>00137 <span class="comment">/* srcBLen is always considered as shorter or equal to srcALen */</span>
155 <a name="l00138"></a>00138 j = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
156 <a name="l00139"></a>00139 srcBLen = <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>;
157 <a name="l00140"></a>00140 srcALen = j;
158 <a name="l00141"></a>00141
159 <a name="l00142"></a>00142 <span class="comment">/* CORR(x, y) = Reverse order(CORR(y, x)) */</span>
160 <a name="l00143"></a>00143 <span class="comment">/* Hence set the destination pointer to point to the last output sample */</span>
161 <a name="l00144"></a>00144 pOut = pDst + ((srcALen + <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>) - 2u);
162 <a name="l00145"></a>00145
163 <a name="l00146"></a>00146 <span class="comment">/* Destination address modifier is set to -1 */</span>
164 <a name="l00147"></a>00147 inc = -1;
165 <a name="l00148"></a>00148
166 <a name="l00149"></a>00149 }
167 <a name="l00150"></a>00150
168 <a name="l00151"></a>00151 <span class="comment">/* The function is internally </span>
169 <a name="l00152"></a>00152 <span class="comment"> * divided into three parts according to the number of multiplications that has to be </span>
170 <a name="l00153"></a>00153 <span class="comment"> * taken place between inputA samples and inputB samples. In the first part of the </span>
171 <a name="l00154"></a>00154 <span class="comment"> * algorithm, the multiplications increase by one for every iteration. </span>
172 <a name="l00155"></a>00155 <span class="comment"> * In the second part of the algorithm, srcBLen number of multiplications are done. </span>
173 <a name="l00156"></a>00156 <span class="comment"> * In the third part of the algorithm, the multiplications decrease by one </span>
174 <a name="l00157"></a>00157 <span class="comment"> * for every iteration.*/</span>
175 <a name="l00158"></a>00158 <span class="comment">/* The algorithm is implemented in three stages. </span>
176 <a name="l00159"></a>00159 <span class="comment"> * The loop counters of each stage is initiated here. */</span>
177 <a name="l00160"></a>00160 blockSize1 = srcBLen - 1u;
178 <a name="l00161"></a>00161 blockSize2 = srcALen - (srcBLen - 1u);
179 <a name="l00162"></a>00162 blockSize3 = blockSize1;
180 <a name="l00163"></a>00163
181 <a name="l00164"></a>00164 <span class="comment">/* -------------------------- </span>
182 <a name="l00165"></a>00165 <span class="comment"> * Initializations of stage1 </span>
183 <a name="l00166"></a>00166 <span class="comment"> * -------------------------*/</span>
184 <a name="l00167"></a>00167
185 <a name="l00168"></a>00168 <span class="comment">/* sum = x[0] * y[srcBlen - 1] </span>
186 <a name="l00169"></a>00169 <span class="comment"> * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1] </span>
187 <a name="l00170"></a>00170 <span class="comment"> * .... </span>
188 <a name="l00171"></a>00171 <span class="comment"> * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] </span>
189 <a name="l00172"></a>00172 <span class="comment"> */</span>
190 <a name="l00173"></a>00173
191 <a name="l00174"></a>00174 <span class="comment">/* In this stage the MAC operations are increased by 1 for every iteration. </span>
192 <a name="l00175"></a>00175 <span class="comment"> The count variable holds the number of MAC operations performed */</span>
193 <a name="l00176"></a>00176 count = 1u;
194 <a name="l00177"></a>00177
195 <a name="l00178"></a>00178 <span class="comment">/* Working pointer of inputA */</span>
196 <a name="l00179"></a>00179 px = pIn1;
197 <a name="l00180"></a>00180
198 <a name="l00181"></a>00181 <span class="comment">/* Working pointer of inputB */</span>
199 <a name="l00182"></a>00182 pSrc1 = pIn2 + (srcBLen - 1u);
200 <a name="l00183"></a>00183 py = pSrc1;
201 <a name="l00184"></a>00184
202 <a name="l00185"></a>00185 <span class="comment">/* ------------------------ </span>
203 <a name="l00186"></a>00186 <span class="comment"> * Stage1 process </span>
204 <a name="l00187"></a>00187 <span class="comment"> * ----------------------*/</span>
205 <a name="l00188"></a>00188
206 <a name="l00189"></a>00189 <span class="comment">/* The first stage starts here */</span>
207 <a name="l00190"></a>00190 <span class="keywordflow">while</span>(blockSize1 > 0u)
208 <a name="l00191"></a>00191 {
209 <a name="l00192"></a>00192 <span class="comment">/* Accumulator is made zero for every iteration */</span>
210 <a name="l00193"></a>00193 sum = 0;
211 <a name="l00194"></a>00194
212 <a name="l00195"></a>00195 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
213 <a name="l00196"></a>00196 k = count >> 2;
214 <a name="l00197"></a>00197
215 <a name="l00198"></a>00198 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
216 <a name="l00199"></a>00199 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
217 <a name="l00200"></a>00200 <span class="keywordflow">while</span>(k > 0u)
218 <a name="l00201"></a>00201 {
219 <a name="l00202"></a>00202 <span class="comment">/* x[0] * y[srcBLen - 4] */</span>
220 <a name="l00203"></a>00203 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
221 <a name="l00204"></a>00204 <span class="comment">/* x[1] * y[srcBLen - 3] */</span>
222 <a name="l00205"></a>00205 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
223 <a name="l00206"></a>00206 <span class="comment">/* x[2] * y[srcBLen - 2] */</span>
224 <a name="l00207"></a>00207 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
225 <a name="l00208"></a>00208 <span class="comment">/* x[3] * y[srcBLen - 1] */</span>
226 <a name="l00209"></a>00209 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
227 <a name="l00210"></a>00210
228 <a name="l00211"></a>00211 <span class="comment">/* Decrement the loop counter */</span>
229 <a name="l00212"></a>00212 k--;
230 <a name="l00213"></a>00213 }
231 <a name="l00214"></a>00214
232 <a name="l00215"></a>00215 <span class="comment">/* If the count is not a multiple of 4, compute any remaining MACs here. </span>
233 <a name="l00216"></a>00216 <span class="comment"> ** No loop unrolling is used. */</span>
234 <a name="l00217"></a>00217 k = count % 0x4u;
235 <a name="l00218"></a>00218
236 <a name="l00219"></a>00219 <span class="keywordflow">while</span>(k > 0u)
237 <a name="l00220"></a>00220 {
238 <a name="l00221"></a>00221 <span class="comment">/* Perform the multiply-accumulates */</span>
239 <a name="l00222"></a>00222 <span class="comment">/* x[0] * y[srcBLen - 1] */</span>
240 <a name="l00223"></a>00223 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
241 <a name="l00224"></a>00224
242 <a name="l00225"></a>00225 <span class="comment">/* Decrement the loop counter */</span>
243 <a name="l00226"></a>00226 k--;
244 <a name="l00227"></a>00227 }
245 <a name="l00228"></a>00228
246 <a name="l00229"></a>00229 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
247 <a name="l00230"></a>00230 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31);
248 <a name="l00231"></a>00231 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
249 <a name="l00232"></a>00232 pOut += inc;
250 <a name="l00233"></a>00233
251 <a name="l00234"></a>00234 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
252 <a name="l00235"></a>00235 py = pSrc1 - count;
253 <a name="l00236"></a>00236 px = pIn1;
254 <a name="l00237"></a>00237
255 <a name="l00238"></a>00238 <span class="comment">/* Increment the MAC count */</span>
256 <a name="l00239"></a>00239 count++;
257 <a name="l00240"></a>00240
258 <a name="l00241"></a>00241 <span class="comment">/* Decrement the loop counter */</span>
259 <a name="l00242"></a>00242 blockSize1--;
260 <a name="l00243"></a>00243 }
261 <a name="l00244"></a>00244
262 <a name="l00245"></a>00245 <span class="comment">/* -------------------------- </span>
263 <a name="l00246"></a>00246 <span class="comment"> * Initializations of stage2 </span>
264 <a name="l00247"></a>00247 <span class="comment"> * ------------------------*/</span>
265 <a name="l00248"></a>00248
266 <a name="l00249"></a>00249 <span class="comment">/* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] </span>
267 <a name="l00250"></a>00250 <span class="comment"> * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] </span>
268 <a name="l00251"></a>00251 <span class="comment"> * .... </span>
269 <a name="l00252"></a>00252 <span class="comment"> * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
270 <a name="l00253"></a>00253 <span class="comment"> */</span>
271 <a name="l00254"></a>00254
272 <a name="l00255"></a>00255 <span class="comment">/* Working pointer of inputA */</span>
273 <a name="l00256"></a>00256 px = pIn1;
274 <a name="l00257"></a>00257
275 <a name="l00258"></a>00258 <span class="comment">/* Working pointer of inputB */</span>
276 <a name="l00259"></a>00259 py = pIn2;
277 <a name="l00260"></a>00260
278 <a name="l00261"></a>00261 <span class="comment">/* count is index by which the pointer pIn1 to be incremented */</span>
279 <a name="l00262"></a>00262 count = 1u;
280 <a name="l00263"></a>00263
281 <a name="l00264"></a>00264 <span class="comment">/* ------------------- </span>
282 <a name="l00265"></a>00265 <span class="comment"> * Stage2 process </span>
283 <a name="l00266"></a>00266 <span class="comment"> * ------------------*/</span>
284 <a name="l00267"></a>00267
285 <a name="l00268"></a>00268 <span class="comment">/* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. </span>
286 <a name="l00269"></a>00269 <span class="comment"> * So, to loop unroll over blockSize2, </span>
287 <a name="l00270"></a>00270 <span class="comment"> * srcBLen should be greater than or equal to 4 */</span>
288 <a name="l00271"></a>00271 <span class="keywordflow">if</span>(srcBLen >= 4u)
289 <a name="l00272"></a>00272 {
290 <a name="l00273"></a>00273 <span class="comment">/* Loop unroll over blockSize2, by 4 */</span>
291 <a name="l00274"></a>00274 blkCnt = blockSize2 >> 2u;
292 <a name="l00275"></a>00275
293 <a name="l00276"></a>00276 <span class="keywordflow">while</span>(blkCnt > 0u)
294 <a name="l00277"></a>00277 {
295 <a name="l00278"></a>00278 <span class="comment">/* Set all accumulators to zero */</span>
296 <a name="l00279"></a>00279 acc0 = 0;
297 <a name="l00280"></a>00280 acc1 = 0;
298 <a name="l00281"></a>00281 acc2 = 0;
299 <a name="l00282"></a>00282 acc3 = 0;
300 <a name="l00283"></a>00283
301 <a name="l00284"></a>00284 <span class="comment">/* read x[0], x[1], x[2] samples */</span>
302 <a name="l00285"></a>00285 x0 = *(px++);
303 <a name="l00286"></a>00286 x1 = *(px++);
304 <a name="l00287"></a>00287 x2 = *(px++);
305 <a name="l00288"></a>00288
306 <a name="l00289"></a>00289 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
307 <a name="l00290"></a>00290 k = srcBLen >> 2u;
308 <a name="l00291"></a>00291
309 <a name="l00292"></a>00292 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
310 <a name="l00293"></a>00293 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
311 <a name="l00294"></a>00294 <span class="keywordflow">do</span>
312 <a name="l00295"></a>00295 {
313 <a name="l00296"></a>00296 <span class="comment">/* Read y[0] sample */</span>
314 <a name="l00297"></a>00297 c0 = *(py++);
315 <a name="l00298"></a>00298
316 <a name="l00299"></a>00299 <span class="comment">/* Read x[3] sample */</span>
317 <a name="l00300"></a>00300 x3 = *(px++);
318 <a name="l00301"></a>00301
319 <a name="l00302"></a>00302 <span class="comment">/* Perform the multiply-accumulate */</span>
320 <a name="l00303"></a>00303 <span class="comment">/* acc0 += x[0] * y[0] */</span>
321 <a name="l00304"></a>00304 acc0 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x0 * c0);
322 <a name="l00305"></a>00305 <span class="comment">/* acc1 += x[1] * y[0] */</span>
323 <a name="l00306"></a>00306 acc1 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x1 * c0);
324 <a name="l00307"></a>00307 <span class="comment">/* acc2 += x[2] * y[0] */</span>
325 <a name="l00308"></a>00308 acc2 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x2 * c0);
326 <a name="l00309"></a>00309 <span class="comment">/* acc3 += x[3] * y[0] */</span>
327 <a name="l00310"></a>00310 acc3 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x3 * c0);
328 <a name="l00311"></a>00311
329 <a name="l00312"></a>00312 <span class="comment">/* Read y[1] sample */</span>
330 <a name="l00313"></a>00313 c0 = *(py++);
331 <a name="l00314"></a>00314
332 <a name="l00315"></a>00315 <span class="comment">/* Read x[4] sample */</span>
333 <a name="l00316"></a>00316 x0 = *(px++);
334 <a name="l00317"></a>00317
335 <a name="l00318"></a>00318 <span class="comment">/* Perform the multiply-accumulates */</span>
336 <a name="l00319"></a>00319 <span class="comment">/* acc0 += x[1] * y[1] */</span>
337 <a name="l00320"></a>00320 acc0 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x1 * c0);
338 <a name="l00321"></a>00321 <span class="comment">/* acc1 += x[2] * y[1] */</span>
339 <a name="l00322"></a>00322 acc1 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x2 * c0);
340 <a name="l00323"></a>00323 <span class="comment">/* acc2 += x[3] * y[1] */</span>
341 <a name="l00324"></a>00324 acc2 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x3 * c0);
342 <a name="l00325"></a>00325 <span class="comment">/* acc3 += x[4] * y[1] */</span>
343 <a name="l00326"></a>00326 acc3 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x0 * c0);
344 <a name="l00327"></a>00327 <span class="comment">/* Read y[2] sample */</span>
345 <a name="l00328"></a>00328 c0 = *(py++);
346 <a name="l00329"></a>00329
347 <a name="l00330"></a>00330 <span class="comment">/* Read x[5] sample */</span>
348 <a name="l00331"></a>00331 x1 = *(px++);
349 <a name="l00332"></a>00332
350 <a name="l00333"></a>00333 <span class="comment">/* Perform the multiply-accumulates */</span>
351 <a name="l00334"></a>00334 <span class="comment">/* acc0 += x[2] * y[2] */</span>
352 <a name="l00335"></a>00335 acc0 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x2 * c0);
353 <a name="l00336"></a>00336 <span class="comment">/* acc1 += x[3] * y[2] */</span>
354 <a name="l00337"></a>00337 acc1 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x3 * c0);
355 <a name="l00338"></a>00338 <span class="comment">/* acc2 += x[4] * y[2] */</span>
356 <a name="l00339"></a>00339 acc2 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x0 * c0);
357 <a name="l00340"></a>00340 <span class="comment">/* acc3 += x[5] * y[2] */</span>
358 <a name="l00341"></a>00341 acc3 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x1 * c0);
359 <a name="l00342"></a>00342
360 <a name="l00343"></a>00343 <span class="comment">/* Read y[3] sample */</span>
361 <a name="l00344"></a>00344 c0 = *(py++);
362 <a name="l00345"></a>00345
363 <a name="l00346"></a>00346 <span class="comment">/* Read x[6] sample */</span>
364 <a name="l00347"></a>00347 x2 = *(px++);
365 <a name="l00348"></a>00348
366 <a name="l00349"></a>00349 <span class="comment">/* Perform the multiply-accumulates */</span>
367 <a name="l00350"></a>00350 <span class="comment">/* acc0 += x[3] * y[3] */</span>
368 <a name="l00351"></a>00351 acc0 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x3 * c0);
369 <a name="l00352"></a>00352 <span class="comment">/* acc1 += x[4] * y[3] */</span>
370 <a name="l00353"></a>00353 acc1 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x0 * c0);
371 <a name="l00354"></a>00354 <span class="comment">/* acc2 += x[5] * y[3] */</span>
372 <a name="l00355"></a>00355 acc2 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x1 * c0);
373 <a name="l00356"></a>00356 <span class="comment">/* acc3 += x[6] * y[3] */</span>
374 <a name="l00357"></a>00357 acc3 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x2 * c0);
375 <a name="l00358"></a>00358
376 <a name="l00359"></a>00359
377 <a name="l00360"></a>00360 } <span class="keywordflow">while</span>(--k);
378 <a name="l00361"></a>00361
379 <a name="l00362"></a>00362 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span>
380 <a name="l00363"></a>00363 <span class="comment"> ** No loop unrolling is used. */</span>
381 <a name="l00364"></a>00364 k = srcBLen % 0x4u;
382 <a name="l00365"></a>00365
383 <a name="l00366"></a>00366 <span class="keywordflow">while</span>(k > 0u)
384 <a name="l00367"></a>00367 {
385 <a name="l00368"></a>00368 <span class="comment">/* Read y[4] sample */</span>
386 <a name="l00369"></a>00369 c0 = *(py++);
387 <a name="l00370"></a>00370
388 <a name="l00371"></a>00371 <span class="comment">/* Read x[7] sample */</span>
389 <a name="l00372"></a>00372 x3 = *(px++);
390 <a name="l00373"></a>00373
391 <a name="l00374"></a>00374 <span class="comment">/* Perform the multiply-accumulates */</span>
392 <a name="l00375"></a>00375 <span class="comment">/* acc0 += x[4] * y[4] */</span>
393 <a name="l00376"></a>00376 acc0 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x0 * c0);
394 <a name="l00377"></a>00377 <span class="comment">/* acc1 += x[5] * y[4] */</span>
395 <a name="l00378"></a>00378 acc1 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x1 * c0);
396 <a name="l00379"></a>00379 <span class="comment">/* acc2 += x[6] * y[4] */</span>
397 <a name="l00380"></a>00380 acc2 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x2 * c0);
398 <a name="l00381"></a>00381 <span class="comment">/* acc3 += x[7] * y[4] */</span>
399 <a name="l00382"></a>00382 acc3 += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) x3 * c0);
400 <a name="l00383"></a>00383
401 <a name="l00384"></a>00384 <span class="comment">/* Reuse the present samples for the next MAC */</span>
402 <a name="l00385"></a>00385 x0 = x1;
403 <a name="l00386"></a>00386 x1 = x2;
404 <a name="l00387"></a>00387 x2 = x3;
405 <a name="l00388"></a>00388
406 <a name="l00389"></a>00389 <span class="comment">/* Decrement the loop counter */</span>
407 <a name="l00390"></a>00390 k--;
408 <a name="l00391"></a>00391 }
409 <a name="l00392"></a>00392
410 <a name="l00393"></a>00393 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
411 <a name="l00394"></a>00394 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (acc0 >> 31);
412 <a name="l00395"></a>00395 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
413 <a name="l00396"></a>00396 pOut += inc;
414 <a name="l00397"></a>00397
415 <a name="l00398"></a>00398 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (acc1 >> 31);
416 <a name="l00399"></a>00399 pOut += inc;
417 <a name="l00400"></a>00400
418 <a name="l00401"></a>00401 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (acc2 >> 31);
419 <a name="l00402"></a>00402 pOut += inc;
420 <a name="l00403"></a>00403
421 <a name="l00404"></a>00404 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (acc3 >> 31);
422 <a name="l00405"></a>00405 pOut += inc;
423 <a name="l00406"></a>00406
424 <a name="l00407"></a>00407 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
425 <a name="l00408"></a>00408 px = pIn1 + (count * 4u);
426 <a name="l00409"></a>00409 py = pIn2;
427 <a name="l00410"></a>00410
428 <a name="l00411"></a>00411 <span class="comment">/* Increment the pointer pIn1 index, count by 1 */</span>
429 <a name="l00412"></a>00412 count++;
430 <a name="l00413"></a>00413
431 <a name="l00414"></a>00414 <span class="comment">/* Decrement the loop counter */</span>
432 <a name="l00415"></a>00415 blkCnt--;
433 <a name="l00416"></a>00416 }
434 <a name="l00417"></a>00417
435 <a name="l00418"></a>00418 <span class="comment">/* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. </span>
436 <a name="l00419"></a>00419 <span class="comment"> ** No loop unrolling is used. */</span>
437 <a name="l00420"></a>00420 blkCnt = blockSize2 % 0x4u;
438 <a name="l00421"></a>00421
439 <a name="l00422"></a>00422 <span class="keywordflow">while</span>(blkCnt > 0u)
440 <a name="l00423"></a>00423 {
441 <a name="l00424"></a>00424 <span class="comment">/* Accumulator is made zero for every iteration */</span>
442 <a name="l00425"></a>00425 sum = 0;
443 <a name="l00426"></a>00426
444 <a name="l00427"></a>00427 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
445 <a name="l00428"></a>00428 k = srcBLen >> 2u;
446 <a name="l00429"></a>00429
447 <a name="l00430"></a>00430 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
448 <a name="l00431"></a>00431 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
449 <a name="l00432"></a>00432 <span class="keywordflow">while</span>(k > 0u)
450 <a name="l00433"></a>00433 {
451 <a name="l00434"></a>00434 <span class="comment">/* Perform the multiply-accumulates */</span>
452 <a name="l00435"></a>00435 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
453 <a name="l00436"></a>00436 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
454 <a name="l00437"></a>00437 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
455 <a name="l00438"></a>00438 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
456 <a name="l00439"></a>00439
457 <a name="l00440"></a>00440 <span class="comment">/* Decrement the loop counter */</span>
458 <a name="l00441"></a>00441 k--;
459 <a name="l00442"></a>00442 }
460 <a name="l00443"></a>00443
461 <a name="l00444"></a>00444 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span>
462 <a name="l00445"></a>00445 <span class="comment"> ** No loop unrolling is used. */</span>
463 <a name="l00446"></a>00446 k = srcBLen % 0x4u;
464 <a name="l00447"></a>00447
465 <a name="l00448"></a>00448 <span class="keywordflow">while</span>(k > 0u)
466 <a name="l00449"></a>00449 {
467 <a name="l00450"></a>00450 <span class="comment">/* Perform the multiply-accumulate */</span>
468 <a name="l00451"></a>00451 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
469 <a name="l00452"></a>00452
470 <a name="l00453"></a>00453 <span class="comment">/* Decrement the loop counter */</span>
471 <a name="l00454"></a>00454 k--;
472 <a name="l00455"></a>00455 }
473 <a name="l00456"></a>00456
474 <a name="l00457"></a>00457 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
475 <a name="l00458"></a>00458 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31);
476 <a name="l00459"></a>00459 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
477 <a name="l00460"></a>00460 pOut += inc;
478 <a name="l00461"></a>00461
479 <a name="l00462"></a>00462 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
480 <a name="l00463"></a>00463 px = pIn1 + count;
481 <a name="l00464"></a>00464 py = pIn2;
482 <a name="l00465"></a>00465
483 <a name="l00466"></a>00466 <span class="comment">/* Increment the MAC count */</span>
484 <a name="l00467"></a>00467 count++;
485 <a name="l00468"></a>00468
486 <a name="l00469"></a>00469 <span class="comment">/* Decrement the loop counter */</span>
487 <a name="l00470"></a>00470 blkCnt--;
488 <a name="l00471"></a>00471 }
489 <a name="l00472"></a>00472 }
490 <a name="l00473"></a>00473 <span class="keywordflow">else</span>
491 <a name="l00474"></a>00474 {
492 <a name="l00475"></a>00475 <span class="comment">/* If the srcBLen is not a multiple of 4, </span>
493 <a name="l00476"></a>00476 <span class="comment"> * the blockSize2 loop cannot be unrolled by 4 */</span>
494 <a name="l00477"></a>00477 blkCnt = blockSize2;
495 <a name="l00478"></a>00478
496 <a name="l00479"></a>00479 <span class="keywordflow">while</span>(blkCnt > 0u)
497 <a name="l00480"></a>00480 {
498 <a name="l00481"></a>00481 <span class="comment">/* Accumulator is made zero for every iteration */</span>
499 <a name="l00482"></a>00482 sum = 0;
500 <a name="l00483"></a>00483
501 <a name="l00484"></a>00484 <span class="comment">/* Loop over srcBLen */</span>
502 <a name="l00485"></a>00485 k = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
503 <a name="l00486"></a>00486
504 <a name="l00487"></a>00487 <span class="keywordflow">while</span>(k > 0u)
505 <a name="l00488"></a>00488 {
506 <a name="l00489"></a>00489 <span class="comment">/* Perform the multiply-accumulate */</span>
507 <a name="l00490"></a>00490 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
508 <a name="l00491"></a>00491
509 <a name="l00492"></a>00492 <span class="comment">/* Decrement the loop counter */</span>
510 <a name="l00493"></a>00493 k--;
511 <a name="l00494"></a>00494 }
512 <a name="l00495"></a>00495
513 <a name="l00496"></a>00496 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
514 <a name="l00497"></a>00497 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31);
515 <a name="l00498"></a>00498 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
516 <a name="l00499"></a>00499 pOut += inc;
517 <a name="l00500"></a>00500
518 <a name="l00501"></a>00501 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
519 <a name="l00502"></a>00502 px = pIn1 + count;
520 <a name="l00503"></a>00503 py = pIn2;
521 <a name="l00504"></a>00504
522 <a name="l00505"></a>00505 <span class="comment">/* Increment the MAC count */</span>
523 <a name="l00506"></a>00506 count++;
524 <a name="l00507"></a>00507
525 <a name="l00508"></a>00508 <span class="comment">/* Decrement the loop counter */</span>
526 <a name="l00509"></a>00509 blkCnt--;
527 <a name="l00510"></a>00510 }
528 <a name="l00511"></a>00511 }
529 <a name="l00512"></a>00512
530 <a name="l00513"></a>00513 <span class="comment">/* -------------------------- </span>
531 <a name="l00514"></a>00514 <span class="comment"> * Initializations of stage3 </span>
532 <a name="l00515"></a>00515 <span class="comment"> * -------------------------*/</span>
533 <a name="l00516"></a>00516
534 <a name="l00517"></a>00517 <span class="comment">/* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
535 <a name="l00518"></a>00518 <span class="comment"> * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
536 <a name="l00519"></a>00519 <span class="comment"> * .... </span>
537 <a name="l00520"></a>00520 <span class="comment"> * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] </span>
538 <a name="l00521"></a>00521 <span class="comment"> * sum += x[srcALen-1] * y[0] </span>
539 <a name="l00522"></a>00522 <span class="comment"> */</span>
540 <a name="l00523"></a>00523
541 <a name="l00524"></a>00524 <span class="comment">/* In this stage the MAC operations are decreased by 1 for every iteration. </span>
542 <a name="l00525"></a>00525 <span class="comment"> The count variable holds the number of MAC operations performed */</span>
543 <a name="l00526"></a>00526 count = srcBLen - 1u;
544 <a name="l00527"></a>00527
545 <a name="l00528"></a>00528 <span class="comment">/* Working pointer of inputA */</span>
546 <a name="l00529"></a>00529 pSrc1 = pIn1 + (srcALen - (srcBLen - 1u));
547 <a name="l00530"></a>00530 px = pSrc1;
548 <a name="l00531"></a>00531
549 <a name="l00532"></a>00532 <span class="comment">/* Working pointer of inputB */</span>
550 <a name="l00533"></a>00533 py = pIn2;
551 <a name="l00534"></a>00534
552 <a name="l00535"></a>00535 <span class="comment">/* ------------------- </span>
553 <a name="l00536"></a>00536 <span class="comment"> * Stage3 process </span>
554 <a name="l00537"></a>00537 <span class="comment"> * ------------------*/</span>
555 <a name="l00538"></a>00538
556 <a name="l00539"></a>00539 <span class="keywordflow">while</span>(blockSize3 > 0u)
557 <a name="l00540"></a>00540 {
558 <a name="l00541"></a>00541 <span class="comment">/* Accumulator is made zero for every iteration */</span>
559 <a name="l00542"></a>00542 sum = 0;
560 <a name="l00543"></a>00543
561 <a name="l00544"></a>00544 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
562 <a name="l00545"></a>00545 k = count >> 2u;
563 <a name="l00546"></a>00546
564 <a name="l00547"></a>00547 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
565 <a name="l00548"></a>00548 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
566 <a name="l00549"></a>00549 <span class="keywordflow">while</span>(k > 0u)
567 <a name="l00550"></a>00550 {
568 <a name="l00551"></a>00551 <span class="comment">/* Perform the multiply-accumulates */</span>
569 <a name="l00552"></a>00552 <span class="comment">/* sum += x[srcALen - srcBLen + 4] * y[3] */</span>
570 <a name="l00553"></a>00553 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
571 <a name="l00554"></a>00554 <span class="comment">/* sum += x[srcALen - srcBLen + 3] * y[2] */</span>
572 <a name="l00555"></a>00555 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
573 <a name="l00556"></a>00556 <span class="comment">/* sum += x[srcALen - srcBLen + 2] * y[1] */</span>
574 <a name="l00557"></a>00557 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
575 <a name="l00558"></a>00558 <span class="comment">/* sum += x[srcALen - srcBLen + 1] * y[0] */</span>
576 <a name="l00559"></a>00559 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
577 <a name="l00560"></a>00560
578 <a name="l00561"></a>00561 <span class="comment">/* Decrement the loop counter */</span>
579 <a name="l00562"></a>00562 k--;
580 <a name="l00563"></a>00563 }
581 <a name="l00564"></a>00564
582 <a name="l00565"></a>00565 <span class="comment">/* If the count is not a multiple of 4, compute any remaining MACs here. </span>
583 <a name="l00566"></a>00566 <span class="comment"> ** No loop unrolling is used. */</span>
584 <a name="l00567"></a>00567 k = count % 0x4u;
585 <a name="l00568"></a>00568
586 <a name="l00569"></a>00569 <span class="keywordflow">while</span>(k > 0u)
587 <a name="l00570"></a>00570 {
588 <a name="l00571"></a>00571 <span class="comment">/* Perform the multiply-accumulates */</span>
589 <a name="l00572"></a>00572 sum += (<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) * px++ * (*py++);
590 <a name="l00573"></a>00573
591 <a name="l00574"></a>00574 <span class="comment">/* Decrement the loop counter */</span>
592 <a name="l00575"></a>00575 k--;
593 <a name="l00576"></a>00576 }
594 <a name="l00577"></a>00577
595 <a name="l00578"></a>00578 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
596 <a name="l00579"></a>00579 *pOut = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31);
597 <a name="l00580"></a>00580 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
598 <a name="l00581"></a>00581 pOut += inc;
599 <a name="l00582"></a>00582
600 <a name="l00583"></a>00583 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
601 <a name="l00584"></a>00584 px = ++pSrc1;
602 <a name="l00585"></a>00585 py = pIn2;
603 <a name="l00586"></a>00586
604 <a name="l00587"></a>00587 <span class="comment">/* Decrement the MAC count */</span>
605 <a name="l00588"></a>00588 count--;
606 <a name="l00589"></a>00589
607 <a name="l00590"></a>00590 <span class="comment">/* Decrement the loop counter */</span>
608 <a name="l00591"></a>00591 blockSize3--;
609 <a name="l00592"></a>00592 }
610 <a name="l00593"></a>00593
611 <a name="l00594"></a>00594 <span class="preprocessor">#else</span>
612 <a name="l00595"></a>00595 <span class="preprocessor"></span>
613 <a name="l00596"></a>00596 <span class="comment">/* Run the below code for Cortex-M0 */</span>
614 <a name="l00597"></a>00597
615 <a name="l00598"></a>00598 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pIn1 = pSrcA; <span class="comment">/* inputA pointer */</span>
616 <a name="l00599"></a>00599 <a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *pIn2 = pSrcB + (srcBLen - 1u); <span class="comment">/* inputB pointer */</span>
617 <a name="l00600"></a>00600 <a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a> sum; <span class="comment">/* Accumulators */</span>
618 <a name="l00601"></a>00601 uint32_t i = 0u, j; <span class="comment">/* loop counters */</span>
619 <a name="l00602"></a>00602 uint32_t inv = 0u; <span class="comment">/* Reverse order flag */</span>
620 <a name="l00603"></a>00603 uint32_t tot = 0u; <span class="comment">/* Length */</span>
621 <a name="l00604"></a>00604
622 <a name="l00605"></a>00605 <span class="comment">/* The algorithm implementation is based on the lengths of the inputs. */</span>
623 <a name="l00606"></a>00606 <span class="comment">/* srcB is always made to slide across srcA. */</span>
624 <a name="l00607"></a>00607 <span class="comment">/* So srcBLen is always considered as shorter or equal to srcALen */</span>
625 <a name="l00608"></a>00608 <span class="comment">/* But CORR(x, y) is reverse of CORR(y, x) */</span>
626 <a name="l00609"></a>00609 <span class="comment">/* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */</span>
627 <a name="l00610"></a>00610 <span class="comment">/* and a varaible, inv is set to 1 */</span>
628 <a name="l00611"></a>00611 <span class="comment">/* If lengths are not equal then zero pad has to be done to make the two </span>
629 <a name="l00612"></a>00612 <span class="comment"> * inputs of same length. But to improve the performance, we include zeroes </span>
630 <a name="l00613"></a>00613 <span class="comment"> * in the output instead of zero padding either of the the inputs*/</span>
631 <a name="l00614"></a>00614 <span class="comment">/* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the </span>
632 <a name="l00615"></a>00615 <span class="comment"> * starting of the output buffer */</span>
633 <a name="l00616"></a>00616 <span class="comment">/* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the </span>
634 <a name="l00617"></a>00617 <span class="comment"> * ending of the output buffer */</span>
635 <a name="l00618"></a>00618 <span class="comment">/* Once the zero padding is done the remaining of the output is calcualted </span>
636 <a name="l00619"></a>00619 <span class="comment"> * using convolution but with the shorter signal time shifted. */</span>
637 <a name="l00620"></a>00620
638 <a name="l00621"></a>00621 <span class="comment">/* Calculate the length of the remaining sequence */</span>
639 <a name="l00622"></a>00622 tot = ((srcALen + <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>) - 2u);
640 <a name="l00623"></a>00623
641 <a name="l00624"></a>00624 <span class="keywordflow">if</span>(srcALen > srcBLen)
642 <a name="l00625"></a>00625 {
643 <a name="l00626"></a>00626 <span class="comment">/* Calculating the number of zeros to be padded to the output */</span>
644 <a name="l00627"></a>00627 j = srcALen - <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
645 <a name="l00628"></a>00628
646 <a name="l00629"></a>00629 <span class="comment">/* Initialise the pointer after zero padding */</span>
647 <a name="l00630"></a>00630 pDst += j;
648 <a name="l00631"></a>00631 }
649 <a name="l00632"></a>00632
650 <a name="l00633"></a>00633 <span class="keywordflow">else</span> <span class="keywordflow">if</span>(srcALen < srcBLen)
651 <a name="l00634"></a>00634 {
652 <a name="l00635"></a>00635 <span class="comment">/* Initialization to inputB pointer */</span>
653 <a name="l00636"></a>00636 pIn1 = pSrcB;
654 <a name="l00637"></a>00637
655 <a name="l00638"></a>00638 <span class="comment">/* Initialization to the end of inputA pointer */</span>
656 <a name="l00639"></a>00639 pIn2 = pSrcA + (srcALen - 1u);
657 <a name="l00640"></a>00640
658 <a name="l00641"></a>00641 <span class="comment">/* Initialisation of the pointer after zero padding */</span>
659 <a name="l00642"></a>00642 pDst = pDst + tot;
660 <a name="l00643"></a>00643
661 <a name="l00644"></a>00644 <span class="comment">/* Swapping the lengths */</span>
662 <a name="l00645"></a>00645 j = <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>;
663 <a name="l00646"></a>00646 srcALen = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
664 <a name="l00647"></a>00647 srcBLen = j;
665 <a name="l00648"></a>00648
666 <a name="l00649"></a>00649 <span class="comment">/* Setting the reverse flag */</span>
667 <a name="l00650"></a>00650 inv = 1;
668 <a name="l00651"></a>00651
669 <a name="l00652"></a>00652 }
670 <a name="l00653"></a>00653
671 <a name="l00654"></a>00654 <span class="comment">/* Loop to calculate convolution for output length number of times */</span>
672 <a name="l00655"></a>00655 <span class="keywordflow">for</span> (i = 0u; i <= tot; i++)
673 <a name="l00656"></a>00656 {
674 <a name="l00657"></a>00657 <span class="comment">/* Initialize sum with zero to carry on MAC operations */</span>
675 <a name="l00658"></a>00658 sum = 0;
676 <a name="l00659"></a>00659
677 <a name="l00660"></a>00660 <span class="comment">/* Loop to perform MAC operations according to convolution equation */</span>
678 <a name="l00661"></a>00661 <span class="keywordflow">for</span> (j = 0u; j <= i; j++)
679 <a name="l00662"></a>00662 {
680 <a name="l00663"></a>00663 <span class="comment">/* Check the array limitations */</span>
681 <a name="l00664"></a>00664 <span class="keywordflow">if</span>((((i - j) < <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>) && (j < srcALen)))
682 <a name="l00665"></a>00665 {
683 <a name="l00666"></a>00666 <span class="comment">/* z[i] += x[i-j] * y[j] */</span>
684 <a name="l00667"></a>00667 sum += ((<a class="code" href="arm__math_8h.html#a5aea1cb12fc02d9d44c8abf217eaa5c6" title="64-bit fractional data type in 1.63 format.">q63_t</a>) pIn1[j] * pIn2[-((int32_t) i - j)]);
685 <a name="l00668"></a>00668 }
686 <a name="l00669"></a>00669 }
687 <a name="l00670"></a>00670 <span class="comment">/* Store the output in the destination buffer */</span>
688 <a name="l00671"></a>00671 <span class="keywordflow">if</span>(inv == 1)
689 <a name="l00672"></a>00672 *pDst-- = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31u);
690 <a name="l00673"></a>00673 <span class="keywordflow">else</span>
691 <a name="l00674"></a>00674 *pDst++ = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) (sum >> 31u);
692 <a name="l00675"></a>00675 }
693 <a name="l00676"></a>00676
694 <a name="l00677"></a>00677 <span class="preprocessor">#endif </span><span class="comment">/* #ifndef ARM_MATH_CM0 */</span>
695 <a name="l00678"></a>00678
696 <a name="l00679"></a>00679 }
697 <a name="l00680"></a>00680
700 <!--- window showing the filter options -->
701 <div id="MSearchSelectWindow"
702 onmouseover="return searchBox.OnSearchSelectShow()"
703 onmouseout="return searchBox.OnSearchSelectHide()"
704 onkeydown="return searchBox.OnSearchSelectKey(event)">
705 <a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(0)"><span class="SelectionMark"> </span>All</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(1)"><span class="SelectionMark"> </span>Data Structures</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(2)"><span class="SelectionMark"> </span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(3)"><span class="SelectionMark"> </span>Functions</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(4)"><span class="SelectionMark"> </span>Variables</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(5)"><span class="SelectionMark"> </span>Typedefs</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(6)"><span class="SelectionMark"> </span>Enumerations</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(7)"><span class="SelectionMark"> </span>Enumerator</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(8)"><span class="SelectionMark"> </span>Defines</a></div>
707 <!-- iframe showing the search results (closed by default) -->
708 <div id="MSearchResultsWindow">
709 <iframe src="" frameborder="0"
710 name="MSearchResults" id="MSearchResults">
714 <hr class="footer"/><address class="footer"><small>Generated on Fri Jul 15 2011 13:16:16 for CMSIS DSP Software Library by 
715 <a href="http://www.doxygen.org/index.html">
716 <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.7.2 </small></address>
projects / fw / stlink / blob