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_fast_q15.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_fast_q15.c</h1> </div>
52 <div class="contents">
53 <a href="arm__correlate__fast__q15_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_fast_q15.c </span>
61 <a name="l00009"></a>00009 <span class="comment">* </span>
62 <a name="l00010"></a>00010 <span class="comment">* Description: Fast Q15 Correlation. </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</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
82 <a name="l00030"></a>00030 <span class="preprocessor">#include "<a class="code" href="arm__math_8h.html">arm_math.h</a>"</span>
83 <a name="l00031"></a>00031
84 <a name="l00066"></a><a class="code" href="group___corr.html#gac8de3da44f58e86c2c86156276ca154f">00066</a> <span class="keywordtype">void</span> <a class="code" href="group___corr.html#gac8de3da44f58e86c2c86156276ca154f" title="Correlation of Q15 sequences (fast version) for Cortex-M3 and Cortex-M4.">arm_correlate_fast_q15</a>(
85 <a name="l00067"></a>00067 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> * pSrcA,
86 <a name="l00068"></a>00068 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>,
87 <a name="l00069"></a>00069 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> * pSrcB,
88 <a name="l00070"></a>00070 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>,
89 <a name="l00071"></a>00071 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> * pDst)
90 <a name="l00072"></a>00072 {
91 <a name="l00073"></a>00073 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *pIn1; <span class="comment">/* inputA pointer */</span>
92 <a name="l00074"></a>00074 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *pIn2; <span class="comment">/* inputB pointer */</span>
93 <a name="l00075"></a>00075 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *pOut = pDst; <span class="comment">/* output pointer */</span>
94 <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> sum, acc0, acc1, acc2, acc3; <span class="comment">/* Accumulators */</span>
95 <a name="l00077"></a>00077 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *px; <span class="comment">/* Intermediate inputA pointer */</span>
96 <a name="l00078"></a>00078 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *py; <span class="comment">/* Intermediate inputB pointer */</span>
97 <a name="l00079"></a>00079 <a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *pSrc1; <span class="comment">/* Intermediate pointers */</span>
98 <a name="l00080"></a>00080 <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>
99 <a name="l00081"></a>00081 uint32_t j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; <span class="comment">/* loop counter */</span>
100 <a name="l00082"></a>00082 int32_t inc = 1; <span class="comment">/* Destination address modifier */</span>
101 <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> *pb; <span class="comment">/* 32 bit pointer for inputB buffer */</span>
102 <a name="l00084"></a>00084
103 <a name="l00085"></a>00085
104 <a name="l00086"></a>00086 <span class="comment">/* The algorithm implementation is based on the lengths of the inputs. */</span>
105 <a name="l00087"></a>00087 <span class="comment">/* srcB is always made to slide across srcA. */</span>
106 <a name="l00088"></a>00088 <span class="comment">/* So srcBLen is always considered as shorter or equal to srcALen */</span>
107 <a name="l00089"></a>00089 <span class="comment">/* But CORR(x, y) is reverse of CORR(y, x) */</span>
108 <a name="l00090"></a>00090 <span class="comment">/* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */</span>
109 <a name="l00091"></a>00091 <span class="comment">/* and the destination pointer modifier, inc is set to -1 */</span>
110 <a name="l00092"></a>00092 <span class="comment">/* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */</span>
111 <a name="l00093"></a>00093 <span class="comment">/* But to improve the performance, </span>
112 <a name="l00094"></a>00094 <span class="comment"> * we include zeroes in the output instead of zero padding either of the the inputs*/</span>
113 <a name="l00095"></a>00095 <span class="comment">/* If srcALen > srcBLen, </span>
114 <a name="l00096"></a>00096 <span class="comment"> * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */</span>
115 <a name="l00097"></a>00097 <span class="comment">/* If srcALen < srcBLen, </span>
116 <a name="l00098"></a>00098 <span class="comment"> * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */</span>
117 <a name="l00099"></a>00099 <span class="keywordflow">if</span>(srcALen >= srcBLen)
118 <a name="l00100"></a>00100 {
119 <a name="l00101"></a>00101 <span class="comment">/* Initialization of inputA pointer */</span>
120 <a name="l00102"></a>00102 pIn1 = (pSrcA);
121 <a name="l00103"></a>00103
122 <a name="l00104"></a>00104 <span class="comment">/* Initialization of inputB pointer */</span>
123 <a name="l00105"></a>00105 pIn2 = (pSrcB);
124 <a name="l00106"></a>00106
125 <a name="l00107"></a>00107 <span class="comment">/* Number of output samples is calculated */</span>
126 <a name="l00108"></a>00108 outBlockSize = (2u * <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>) - 1u;
127 <a name="l00109"></a>00109
128 <a name="l00110"></a>00110 <span class="comment">/* When srcALen > srcBLen, zero padding is done to srcB </span>
129 <a name="l00111"></a>00111 <span class="comment"> * to make their lengths equal. </span>
130 <a name="l00112"></a>00112 <span class="comment"> * Instead, (outBlockSize - (srcALen + srcBLen - 1)) </span>
131 <a name="l00113"></a>00113 <span class="comment"> * number of output samples are made zero */</span>
132 <a name="l00114"></a>00114 j = outBlockSize - (srcALen + (srcBLen - 1u));
133 <a name="l00115"></a>00115
134 <a name="l00116"></a>00116 <span class="comment">/* Updating the pointer position to non zero value */</span>
135 <a name="l00117"></a>00117 pOut += j;
136 <a name="l00118"></a>00118
137 <a name="l00119"></a>00119 }
138 <a name="l00120"></a>00120 <span class="keywordflow">else</span>
139 <a name="l00121"></a>00121 {
140 <a name="l00122"></a>00122 <span class="comment">/* Initialization of inputA pointer */</span>
141 <a name="l00123"></a>00123 pIn1 = (pSrcB);
142 <a name="l00124"></a>00124
143 <a name="l00125"></a>00125 <span class="comment">/* Initialization of inputB pointer */</span>
144 <a name="l00126"></a>00126 pIn2 = (pSrcA);
145 <a name="l00127"></a>00127
146 <a name="l00128"></a>00128 <span class="comment">/* srcBLen is always considered as shorter or equal to srcALen */</span>
147 <a name="l00129"></a>00129 j = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
148 <a name="l00130"></a>00130 srcBLen = <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>;
149 <a name="l00131"></a>00131 srcALen = j;
150 <a name="l00132"></a>00132
151 <a name="l00133"></a>00133 <span class="comment">/* CORR(x, y) = Reverse order(CORR(y, x)) */</span>
152 <a name="l00134"></a>00134 <span class="comment">/* Hence set the destination pointer to point to the last output sample */</span>
153 <a name="l00135"></a>00135 pOut = pDst + ((srcALen + <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>) - 2u);
154 <a name="l00136"></a>00136
155 <a name="l00137"></a>00137 <span class="comment">/* Destination address modifier is set to -1 */</span>
156 <a name="l00138"></a>00138 inc = -1;
157 <a name="l00139"></a>00139
158 <a name="l00140"></a>00140 }
159 <a name="l00141"></a>00141
160 <a name="l00142"></a>00142 <span class="comment">/* The function is internally </span>
161 <a name="l00143"></a>00143 <span class="comment"> * divided into three parts according to the number of multiplications that has to be </span>
162 <a name="l00144"></a>00144 <span class="comment"> * taken place between inputA samples and inputB samples. In the first part of the </span>
163 <a name="l00145"></a>00145 <span class="comment"> * algorithm, the multiplications increase by one for every iteration. </span>
164 <a name="l00146"></a>00146 <span class="comment"> * In the second part of the algorithm, srcBLen number of multiplications are done. </span>
165 <a name="l00147"></a>00147 <span class="comment"> * In the third part of the algorithm, the multiplications decrease by one </span>
166 <a name="l00148"></a>00148 <span class="comment"> * for every iteration.*/</span>
167 <a name="l00149"></a>00149 <span class="comment">/* The algorithm is implemented in three stages. </span>
168 <a name="l00150"></a>00150 <span class="comment"> * The loop counters of each stage is initiated here. */</span>
169 <a name="l00151"></a>00151 blockSize1 = srcBLen - 1u;
170 <a name="l00152"></a>00152 blockSize2 = srcALen - (srcBLen - 1u);
171 <a name="l00153"></a>00153 blockSize3 = blockSize1;
172 <a name="l00154"></a>00154
173 <a name="l00155"></a>00155 <span class="comment">/* -------------------------- </span>
174 <a name="l00156"></a>00156 <span class="comment"> * Initializations of stage1 </span>
175 <a name="l00157"></a>00157 <span class="comment"> * -------------------------*/</span>
176 <a name="l00158"></a>00158
177 <a name="l00159"></a>00159 <span class="comment">/* sum = x[0] * y[srcBlen - 1] </span>
178 <a name="l00160"></a>00160 <span class="comment"> * sum = x[0] * y[srcBlen - 2] + x[1] * y[srcBlen - 1] </span>
179 <a name="l00161"></a>00161 <span class="comment"> * .... </span>
180 <a name="l00162"></a>00162 <span class="comment"> * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] </span>
181 <a name="l00163"></a>00163 <span class="comment"> */</span>
182 <a name="l00164"></a>00164
183 <a name="l00165"></a>00165 <span class="comment">/* In this stage the MAC operations are increased by 1 for every iteration. </span>
184 <a name="l00166"></a>00166 <span class="comment"> The count variable holds the number of MAC operations performed */</span>
185 <a name="l00167"></a>00167 count = 1u;
186 <a name="l00168"></a>00168
187 <a name="l00169"></a>00169 <span class="comment">/* Working pointer of inputA */</span>
188 <a name="l00170"></a>00170 px = pIn1;
189 <a name="l00171"></a>00171
190 <a name="l00172"></a>00172 <span class="comment">/* Working pointer of inputB */</span>
191 <a name="l00173"></a>00173 pSrc1 = pIn2 + (srcBLen - 1u);
192 <a name="l00174"></a>00174 py = pSrc1;
193 <a name="l00175"></a>00175
194 <a name="l00176"></a>00176 <span class="comment">/* ------------------------ </span>
195 <a name="l00177"></a>00177 <span class="comment"> * Stage1 process </span>
196 <a name="l00178"></a>00178 <span class="comment"> * ----------------------*/</span>
197 <a name="l00179"></a>00179
198 <a name="l00180"></a>00180 <span class="comment">/* The first loop starts here */</span>
199 <a name="l00181"></a>00181 <span class="keywordflow">while</span>(blockSize1 > 0u)
200 <a name="l00182"></a>00182 {
201 <a name="l00183"></a>00183 <span class="comment">/* Accumulator is made zero for every iteration */</span>
202 <a name="l00184"></a>00184 sum = 0;
203 <a name="l00185"></a>00185
204 <a name="l00186"></a>00186 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
205 <a name="l00187"></a>00187 k = count >> 2;
206 <a name="l00188"></a>00188
207 <a name="l00189"></a>00189 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
208 <a name="l00190"></a>00190 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
209 <a name="l00191"></a>00191 <span class="keywordflow">while</span>(k > 0u)
210 <a name="l00192"></a>00192 {
211 <a name="l00193"></a>00193 <span class="comment">/* x[0] * y[srcBLen - 4] , x[1] * y[srcBLen - 3] */</span>
212 <a name="l00194"></a>00194 sum = __SMLAD(*<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(px)++, *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(py)++, sum);
213 <a name="l00195"></a>00195 <span class="comment">/* x[3] * y[srcBLen - 1] , x[2] * y[srcBLen - 2] */</span>
214 <a name="l00196"></a>00196 sum = __SMLAD(*<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(px)++, *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(py)++, sum);
215 <a name="l00197"></a>00197
216 <a name="l00198"></a>00198 <span class="comment">/* Decrement the loop counter */</span>
217 <a name="l00199"></a>00199 k--;
218 <a name="l00200"></a>00200 }
219 <a name="l00201"></a>00201
220 <a name="l00202"></a>00202 <span class="comment">/* If the count is not a multiple of 4, compute any remaining MACs here. </span>
221 <a name="l00203"></a>00203 <span class="comment"> ** No loop unrolling is used. */</span>
222 <a name="l00204"></a>00204 k = count % 0x4u;
223 <a name="l00205"></a>00205
224 <a name="l00206"></a>00206 <span class="keywordflow">while</span>(k > 0u)
225 <a name="l00207"></a>00207 {
226 <a name="l00208"></a>00208 <span class="comment">/* Perform the multiply-accumulates */</span>
227 <a name="l00209"></a>00209 <span class="comment">/* x[0] * y[srcBLen - 1] */</span>
228 <a name="l00210"></a>00210 sum = __SMLAD(*px++, *py++, sum);
229 <a name="l00211"></a>00211
230 <a name="l00212"></a>00212 <span class="comment">/* Decrement the loop counter */</span>
231 <a name="l00213"></a>00213 k--;
232 <a name="l00214"></a>00214 }
233 <a name="l00215"></a>00215
234 <a name="l00216"></a>00216 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
235 <a name="l00217"></a>00217 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (sum >> 15);
236 <a name="l00218"></a>00218 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
237 <a name="l00219"></a>00219 pOut += inc;
238 <a name="l00220"></a>00220
239 <a name="l00221"></a>00221 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
240 <a name="l00222"></a>00222 py = pSrc1 - count;
241 <a name="l00223"></a>00223 px = pIn1;
242 <a name="l00224"></a>00224
243 <a name="l00225"></a>00225 <span class="comment">/* Increment the MAC count */</span>
244 <a name="l00226"></a>00226 count++;
245 <a name="l00227"></a>00227
246 <a name="l00228"></a>00228 <span class="comment">/* Decrement the loop counter */</span>
247 <a name="l00229"></a>00229 blockSize1--;
248 <a name="l00230"></a>00230 }
249 <a name="l00231"></a>00231
250 <a name="l00232"></a>00232 <span class="comment">/* -------------------------- </span>
251 <a name="l00233"></a>00233 <span class="comment"> * Initializations of stage2 </span>
252 <a name="l00234"></a>00234 <span class="comment"> * ------------------------*/</span>
253 <a name="l00235"></a>00235
254 <a name="l00236"></a>00236 <span class="comment">/* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] </span>
255 <a name="l00237"></a>00237 <span class="comment"> * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] </span>
256 <a name="l00238"></a>00238 <span class="comment"> * .... </span>
257 <a name="l00239"></a>00239 <span class="comment"> * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
258 <a name="l00240"></a>00240 <span class="comment"> */</span>
259 <a name="l00241"></a>00241
260 <a name="l00242"></a>00242 <span class="comment">/* Working pointer of inputA */</span>
261 <a name="l00243"></a>00243 px = pIn1;
262 <a name="l00244"></a>00244
263 <a name="l00245"></a>00245 <span class="comment">/* Working pointer of inputB */</span>
264 <a name="l00246"></a>00246 py = pIn2;
265 <a name="l00247"></a>00247
266 <a name="l00248"></a>00248 <span class="comment">/* Initialize inputB pointer of type q31 */</span>
267 <a name="l00249"></a>00249 pb = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (py);
268 <a name="l00250"></a>00250
269 <a name="l00251"></a>00251 <span class="comment">/* count is index by which the pointer pIn1 to be incremented */</span>
270 <a name="l00252"></a>00252 count = 0u;
271 <a name="l00253"></a>00253
272 <a name="l00254"></a>00254 <span class="comment">/* ------------------- </span>
273 <a name="l00255"></a>00255 <span class="comment"> * Stage2 process </span>
274 <a name="l00256"></a>00256 <span class="comment"> * ------------------*/</span>
275 <a name="l00257"></a>00257
276 <a name="l00258"></a>00258 <span class="comment">/* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. </span>
277 <a name="l00259"></a>00259 <span class="comment"> * So, to loop unroll over blockSize2, </span>
278 <a name="l00260"></a>00260 <span class="comment"> * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */</span>
279 <a name="l00261"></a>00261 <span class="keywordflow">if</span>(srcBLen >= 4u)
280 <a name="l00262"></a>00262 {
281 <a name="l00263"></a>00263 <span class="comment">/* Loop unroll over blockSize2, by 4 */</span>
282 <a name="l00264"></a>00264 blkCnt = blockSize2 >> 2u;
283 <a name="l00265"></a>00265
284 <a name="l00266"></a>00266 <span class="keywordflow">while</span>(blkCnt > 0u)
285 <a name="l00267"></a>00267 {
286 <a name="l00268"></a>00268 <span class="comment">/* Set all accumulators to zero */</span>
287 <a name="l00269"></a>00269 acc0 = 0;
288 <a name="l00270"></a>00270 acc1 = 0;
289 <a name="l00271"></a>00271 acc2 = 0;
290 <a name="l00272"></a>00272 acc3 = 0;
291 <a name="l00273"></a>00273
292 <a name="l00274"></a>00274 <span class="comment">/* read x[0], x[1] samples */</span>
293 <a name="l00275"></a>00275 x0 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
294 <a name="l00276"></a>00276 <span class="comment">/* read x[1], x[2] samples */</span>
295 <a name="l00277"></a>00277 x1 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
296 <a name="l00278"></a>00278
297 <a name="l00279"></a>00279 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
298 <a name="l00280"></a>00280 k = srcBLen >> 2u;
299 <a name="l00281"></a>00281
300 <a name="l00282"></a>00282 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
301 <a name="l00283"></a>00283 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
302 <a name="l00284"></a>00284 <span class="keywordflow">do</span>
303 <a name="l00285"></a>00285 {
304 <a name="l00286"></a>00286 <span class="comment">/* Read the first two inputB samples using SIMD: </span>
305 <a name="l00287"></a>00287 <span class="comment"> * y[0] and y[1] */</span>
306 <a name="l00288"></a>00288 c0 = *(pb++);
307 <a name="l00289"></a>00289
308 <a name="l00290"></a>00290 <span class="comment">/* acc0 += x[0] * y[0] + x[1] * y[1] */</span>
309 <a name="l00291"></a>00291 acc0 = __SMLAD(x0, c0, acc0);
310 <a name="l00292"></a>00292
311 <a name="l00293"></a>00293 <span class="comment">/* acc1 += x[1] * y[0] + x[2] * y[1] */</span>
312 <a name="l00294"></a>00294 acc1 = __SMLAD(x1, c0, acc1);
313 <a name="l00295"></a>00295
314 <a name="l00296"></a>00296 <span class="comment">/* Read x[2], x[3] */</span>
315 <a name="l00297"></a>00297 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
316 <a name="l00298"></a>00298
317 <a name="l00299"></a>00299 <span class="comment">/* Read x[3], x[4] */</span>
318 <a name="l00300"></a>00300 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
319 <a name="l00301"></a>00301
320 <a name="l00302"></a>00302 <span class="comment">/* acc2 += x[2] * y[0] + x[3] * y[1] */</span>
321 <a name="l00303"></a>00303 acc2 = __SMLAD(x2, c0, acc2);
322 <a name="l00304"></a>00304
323 <a name="l00305"></a>00305 <span class="comment">/* acc3 += x[3] * y[0] + x[4] * y[1] */</span>
324 <a name="l00306"></a>00306 acc3 = __SMLAD(x3, c0, acc3);
325 <a name="l00307"></a>00307
326 <a name="l00308"></a>00308 <span class="comment">/* Read y[2] and y[3] */</span>
327 <a name="l00309"></a>00309 c0 = *(pb++);
328 <a name="l00310"></a>00310
329 <a name="l00311"></a>00311 <span class="comment">/* acc0 += x[2] * y[2] + x[3] * y[3] */</span>
330 <a name="l00312"></a>00312 acc0 = __SMLAD(x2, c0, acc0);
331 <a name="l00313"></a>00313
332 <a name="l00314"></a>00314 <span class="comment">/* acc1 += x[3] * y[2] + x[4] * y[3] */</span>
333 <a name="l00315"></a>00315 acc1 = __SMLAD(x3, c0, acc1);
334 <a name="l00316"></a>00316
335 <a name="l00317"></a>00317 <span class="comment">/* Read x[4], x[5] */</span>
336 <a name="l00318"></a>00318 x0 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
337 <a name="l00319"></a>00319
338 <a name="l00320"></a>00320 <span class="comment">/* Read x[5], x[6] */</span>
339 <a name="l00321"></a>00321 x1 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++);
340 <a name="l00322"></a>00322
341 <a name="l00323"></a>00323 <span class="comment">/* acc2 += x[4] * y[2] + x[5] * y[3] */</span>
342 <a name="l00324"></a>00324 acc2 = __SMLAD(x0, c0, acc2);
343 <a name="l00325"></a>00325
344 <a name="l00326"></a>00326 <span class="comment">/* acc3 += x[5] * y[2] + x[6] * y[3] */</span>
345 <a name="l00327"></a>00327 acc3 = __SMLAD(x1, c0, acc3);
346 <a name="l00328"></a>00328
347 <a name="l00329"></a>00329 } <span class="keywordflow">while</span>(--k);
348 <a name="l00330"></a>00330
349 <a name="l00331"></a>00331 <span class="comment">/* For the next MAC operations, SIMD is not used </span>
350 <a name="l00332"></a>00332 <span class="comment"> * So, the 16 bit pointer if inputB, py is updated */</span>
351 <a name="l00333"></a>00333 py = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *) (pb);
352 <a name="l00334"></a>00334
353 <a name="l00335"></a>00335 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span>
354 <a name="l00336"></a>00336 <span class="comment"> ** No loop unrolling is used. */</span>
355 <a name="l00337"></a>00337 k = srcBLen % 0x4u;
356 <a name="l00338"></a>00338
357 <a name="l00339"></a>00339 <span class="keywordflow">if</span>(k == 1u)
358 <a name="l00340"></a>00340 {
359 <a name="l00341"></a>00341 <span class="comment">/* Read y[4] */</span>
360 <a name="l00342"></a>00342 c0 = *py;
361 <a name="l00343"></a>00343 <span class="preprocessor">#ifdef ARM_MATH_BIG_ENDIAN</span>
362 <a name="l00344"></a>00344 <span class="preprocessor"></span>
363 <a name="l00345"></a>00345 c0 = c0 << 16u;
364 <a name="l00346"></a>00346
365 <a name="l00347"></a>00347 <span class="preprocessor">#else</span>
366 <a name="l00348"></a>00348 <span class="preprocessor"></span>
367 <a name="l00349"></a>00349 c0 = c0 & 0x0000FFFF;
368 <a name="l00350"></a>00350
369 <a name="l00351"></a>00351 <span class="preprocessor">#endif </span><span class="comment">/* #ifdef ARM_MATH_BIG_ENDIAN */</span>
370 <a name="l00352"></a>00352
371 <a name="l00353"></a>00353 <span class="comment">/* Read x[7] */</span>
372 <a name="l00354"></a>00354 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
373 <a name="l00355"></a>00355
374 <a name="l00356"></a>00356 <span class="comment">/* Perform the multiply-accumulates */</span>
375 <a name="l00357"></a>00357 acc0 = __SMLAD(x0, c0, acc0);
376 <a name="l00358"></a>00358 acc1 = __SMLAD(x1, c0, acc1);
377 <a name="l00359"></a>00359 acc2 = __SMLADX(x1, c0, acc2);
378 <a name="l00360"></a>00360 acc3 = __SMLADX(x3, c0, acc3);
379 <a name="l00361"></a>00361 }
380 <a name="l00362"></a>00362
381 <a name="l00363"></a>00363 <span class="keywordflow">if</span>(k == 2u)
382 <a name="l00364"></a>00364 {
383 <a name="l00365"></a>00365 <span class="comment">/* Read y[4], y[5] */</span>
384 <a name="l00366"></a>00366 c0 = *(pb);
385 <a name="l00367"></a>00367
386 <a name="l00368"></a>00368 <span class="comment">/* Read x[7], x[8] */</span>
387 <a name="l00369"></a>00369 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
388 <a name="l00370"></a>00370
389 <a name="l00371"></a>00371 <span class="comment">/* Read x[9] */</span>
390 <a name="l00372"></a>00372 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
391 <a name="l00373"></a>00373
392 <a name="l00374"></a>00374 <span class="comment">/* Perform the multiply-accumulates */</span>
393 <a name="l00375"></a>00375 acc0 = __SMLAD(x0, c0, acc0);
394 <a name="l00376"></a>00376 acc1 = __SMLAD(x1, c0, acc1);
395 <a name="l00377"></a>00377 acc2 = __SMLAD(x3, c0, acc2);
396 <a name="l00378"></a>00378 acc3 = __SMLAD(x2, c0, acc3);
397 <a name="l00379"></a>00379 }
398 <a name="l00380"></a>00380
399 <a name="l00381"></a>00381 <span class="keywordflow">if</span>(k == 3u)
400 <a name="l00382"></a>00382 {
401 <a name="l00383"></a>00383 <span class="comment">/* Read y[4], y[5] */</span>
402 <a name="l00384"></a>00384 c0 = *pb++;
403 <a name="l00385"></a>00385
404 <a name="l00386"></a>00386 <span class="comment">/* Read x[7], x[8] */</span>
405 <a name="l00387"></a>00387 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
406 <a name="l00388"></a>00388
407 <a name="l00389"></a>00389 <span class="comment">/* Read x[9] */</span>
408 <a name="l00390"></a>00390 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
409 <a name="l00391"></a>00391
410 <a name="l00392"></a>00392 <span class="comment">/* Perform the multiply-accumulates */</span>
411 <a name="l00393"></a>00393 acc0 = __SMLAD(x0, c0, acc0);
412 <a name="l00394"></a>00394 acc1 = __SMLAD(x1, c0, acc1);
413 <a name="l00395"></a>00395 acc2 = __SMLAD(x3, c0, acc2);
414 <a name="l00396"></a>00396 acc3 = __SMLAD(x2, c0, acc3);
415 <a name="l00397"></a>00397
416 <a name="l00398"></a>00398 <span class="comment">/* Read y[6] */</span>
417 <a name="l00399"></a>00399 <span class="preprocessor">#ifdef ARM_MATH_BIG_ENDIAN</span>
418 <a name="l00400"></a>00400 <span class="preprocessor"></span> c0 = (*pb);
419 <a name="l00401"></a>00401 c0 = c0 & 0xFFFF0000;
420 <a name="l00402"></a>00402
421 <a name="l00403"></a>00403 <span class="preprocessor">#else</span>
422 <a name="l00404"></a>00404 <span class="preprocessor"></span> c0 = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (*pb);
423 <a name="l00405"></a>00405 c0 = c0 & 0x0000FFFF;
424 <a name="l00406"></a>00406
425 <a name="l00407"></a>00407 <span class="preprocessor">#endif </span><span class="comment">/* #ifdef ARM_MATH_BIG_ENDIAN */</span>
426 <a name="l00408"></a>00408
427 <a name="l00409"></a>00409 <span class="comment">/* Read x[10] */</span>
428 <a name="l00410"></a>00410 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++;
429 <a name="l00411"></a>00411
430 <a name="l00412"></a>00412 <span class="comment">/* Perform the multiply-accumulates */</span>
431 <a name="l00413"></a>00413 acc0 = __SMLADX(x1, c0, acc0);
432 <a name="l00414"></a>00414 acc1 = __SMLAD(x2, c0, acc1);
433 <a name="l00415"></a>00415 acc2 = __SMLADX(x2, c0, acc2);
434 <a name="l00416"></a>00416 acc3 = __SMLADX(x3, c0, acc3);
435 <a name="l00417"></a>00417 }
436 <a name="l00418"></a>00418
437 <a name="l00419"></a>00419 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
438 <a name="l00420"></a>00420 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (acc0 >> 15);
439 <a name="l00421"></a>00421 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
440 <a name="l00422"></a>00422 pOut += inc;
441 <a name="l00423"></a>00423
442 <a name="l00424"></a>00424 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (acc1 >> 15);
443 <a name="l00425"></a>00425 pOut += inc;
444 <a name="l00426"></a>00426
445 <a name="l00427"></a>00427 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (acc2 >> 15);
446 <a name="l00428"></a>00428 pOut += inc;
447 <a name="l00429"></a>00429
448 <a name="l00430"></a>00430 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (acc3 >> 15);
449 <a name="l00431"></a>00431 pOut += inc;
450 <a name="l00432"></a>00432
451 <a name="l00433"></a>00433 <span class="comment">/* Increment the pointer pIn1 index, count by 1 */</span>
452 <a name="l00434"></a>00434 count += 4u;
453 <a name="l00435"></a>00435
454 <a name="l00436"></a>00436 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
455 <a name="l00437"></a>00437 px = pIn1 + count;
456 <a name="l00438"></a>00438 py = pIn2;
457 <a name="l00439"></a>00439 pb = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (py);
458 <a name="l00440"></a>00440
459 <a name="l00441"></a>00441
460 <a name="l00442"></a>00442 <span class="comment">/* Decrement the loop counter */</span>
461 <a name="l00443"></a>00443 blkCnt--;
462 <a name="l00444"></a>00444 }
463 <a name="l00445"></a>00445
464 <a name="l00446"></a>00446 <span class="comment">/* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. </span>
465 <a name="l00447"></a>00447 <span class="comment"> ** No loop unrolling is used. */</span>
466 <a name="l00448"></a>00448 blkCnt = blockSize2 % 0x4u;
467 <a name="l00449"></a>00449
468 <a name="l00450"></a>00450 <span class="keywordflow">while</span>(blkCnt > 0u)
469 <a name="l00451"></a>00451 {
470 <a name="l00452"></a>00452 <span class="comment">/* Accumulator is made zero for every iteration */</span>
471 <a name="l00453"></a>00453 sum = 0;
472 <a name="l00454"></a>00454
473 <a name="l00455"></a>00455 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
474 <a name="l00456"></a>00456 k = srcBLen >> 2u;
475 <a name="l00457"></a>00457
476 <a name="l00458"></a>00458 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
477 <a name="l00459"></a>00459 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
478 <a name="l00460"></a>00460 <span class="keywordflow">while</span>(k > 0u)
479 <a name="l00461"></a>00461 {
480 <a name="l00462"></a>00462 <span class="comment">/* Perform the multiply-accumulates */</span>
481 <a name="l00463"></a>00463 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
482 <a name="l00464"></a>00464 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
483 <a name="l00465"></a>00465 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
484 <a name="l00466"></a>00466 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
485 <a name="l00467"></a>00467
486 <a name="l00468"></a>00468 <span class="comment">/* Decrement the loop counter */</span>
487 <a name="l00469"></a>00469 k--;
488 <a name="l00470"></a>00470 }
489 <a name="l00471"></a>00471
490 <a name="l00472"></a>00472 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span>
491 <a name="l00473"></a>00473 <span class="comment"> ** No loop unrolling is used. */</span>
492 <a name="l00474"></a>00474 k = srcBLen % 0x4u;
493 <a name="l00475"></a>00475
494 <a name="l00476"></a>00476 <span class="keywordflow">while</span>(k > 0u)
495 <a name="l00477"></a>00477 {
496 <a name="l00478"></a>00478 <span class="comment">/* Perform the multiply-accumulates */</span>
497 <a name="l00479"></a>00479 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
498 <a name="l00480"></a>00480
499 <a name="l00481"></a>00481 <span class="comment">/* Decrement the loop counter */</span>
500 <a name="l00482"></a>00482 k--;
501 <a name="l00483"></a>00483 }
502 <a name="l00484"></a>00484
503 <a name="l00485"></a>00485 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
504 <a name="l00486"></a>00486 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (sum >> 15);
505 <a name="l00487"></a>00487 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
506 <a name="l00488"></a>00488 pOut += inc;
507 <a name="l00489"></a>00489
508 <a name="l00490"></a>00490 <span class="comment">/* Increment the pointer pIn1 index, count by 1 */</span>
509 <a name="l00491"></a>00491 count++;
510 <a name="l00492"></a>00492
511 <a name="l00493"></a>00493 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
512 <a name="l00494"></a>00494 px = pIn1 + count;
513 <a name="l00495"></a>00495 py = pIn2;
514 <a name="l00496"></a>00496
515 <a name="l00497"></a>00497 <span class="comment">/* Decrement the loop counter */</span>
516 <a name="l00498"></a>00498 blkCnt--;
517 <a name="l00499"></a>00499 }
518 <a name="l00500"></a>00500 }
519 <a name="l00501"></a>00501 <span class="keywordflow">else</span>
520 <a name="l00502"></a>00502 {
521 <a name="l00503"></a>00503 <span class="comment">/* If the srcBLen is not a multiple of 4, </span>
522 <a name="l00504"></a>00504 <span class="comment"> * the blockSize2 loop cannot be unrolled by 4 */</span>
523 <a name="l00505"></a>00505 blkCnt = blockSize2;
524 <a name="l00506"></a>00506
525 <a name="l00507"></a>00507 <span class="keywordflow">while</span>(blkCnt > 0u)
526 <a name="l00508"></a>00508 {
527 <a name="l00509"></a>00509 <span class="comment">/* Accumulator is made zero for every iteration */</span>
528 <a name="l00510"></a>00510 sum = 0;
529 <a name="l00511"></a>00511
530 <a name="l00512"></a>00512 <span class="comment">/* Loop over srcBLen */</span>
531 <a name="l00513"></a>00513 k = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>;
532 <a name="l00514"></a>00514
533 <a name="l00515"></a>00515 <span class="keywordflow">while</span>(k > 0u)
534 <a name="l00516"></a>00516 {
535 <a name="l00517"></a>00517 <span class="comment">/* Perform the multiply-accumulate */</span>
536 <a name="l00518"></a>00518 sum += ((<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a>) * px++ * *py++);
537 <a name="l00519"></a>00519
538 <a name="l00520"></a>00520 <span class="comment">/* Decrement the loop counter */</span>
539 <a name="l00521"></a>00521 k--;
540 <a name="l00522"></a>00522 }
541 <a name="l00523"></a>00523
542 <a name="l00524"></a>00524 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
543 <a name="l00525"></a>00525 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (sum >> 15);
544 <a name="l00526"></a>00526 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
545 <a name="l00527"></a>00527 pOut += inc;
546 <a name="l00528"></a>00528
547 <a name="l00529"></a>00529 <span class="comment">/* Increment the MAC count */</span>
548 <a name="l00530"></a>00530 count++;
549 <a name="l00531"></a>00531
550 <a name="l00532"></a>00532 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
551 <a name="l00533"></a>00533 px = pIn1 + count;
552 <a name="l00534"></a>00534 py = pIn2;
553 <a name="l00535"></a>00535
554 <a name="l00536"></a>00536 <span class="comment">/* Decrement the loop counter */</span>
555 <a name="l00537"></a>00537 blkCnt--;
556 <a name="l00538"></a>00538 }
557 <a name="l00539"></a>00539 }
558 <a name="l00540"></a>00540
559 <a name="l00541"></a>00541 <span class="comment">/* -------------------------- </span>
560 <a name="l00542"></a>00542 <span class="comment"> * Initializations of stage3 </span>
561 <a name="l00543"></a>00543 <span class="comment"> * -------------------------*/</span>
562 <a name="l00544"></a>00544
563 <a name="l00545"></a>00545 <span class="comment">/* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
564 <a name="l00546"></a>00546 <span class="comment"> * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] </span>
565 <a name="l00547"></a>00547 <span class="comment"> * .... </span>
566 <a name="l00548"></a>00548 <span class="comment"> * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] </span>
567 <a name="l00549"></a>00549 <span class="comment"> * sum += x[srcALen-1] * y[0] </span>
568 <a name="l00550"></a>00550 <span class="comment"> */</span>
569 <a name="l00551"></a>00551
570 <a name="l00552"></a>00552 <span class="comment">/* In this stage the MAC operations are decreased by 1 for every iteration. </span>
571 <a name="l00553"></a>00553 <span class="comment"> The count variable holds the number of MAC operations performed */</span>
572 <a name="l00554"></a>00554 count = srcBLen - 1u;
573 <a name="l00555"></a>00555
574 <a name="l00556"></a>00556 <span class="comment">/* Working pointer of inputA */</span>
575 <a name="l00557"></a>00557 pSrc1 = (pIn1 + <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>) - (srcBLen - 1u);
576 <a name="l00558"></a>00558 px = pSrc1;
577 <a name="l00559"></a>00559
578 <a name="l00560"></a>00560 <span class="comment">/* Working pointer of inputB */</span>
579 <a name="l00561"></a>00561 py = pIn2;
580 <a name="l00562"></a>00562
581 <a name="l00563"></a>00563 <span class="comment">/* ------------------- </span>
582 <a name="l00564"></a>00564 <span class="comment"> * Stage3 process </span>
583 <a name="l00565"></a>00565 <span class="comment"> * ------------------*/</span>
584 <a name="l00566"></a>00566
585 <a name="l00567"></a>00567 <span class="keywordflow">while</span>(blockSize3 > 0u)
586 <a name="l00568"></a>00568 {
587 <a name="l00569"></a>00569 <span class="comment">/* Accumulator is made zero for every iteration */</span>
588 <a name="l00570"></a>00570 sum = 0;
589 <a name="l00571"></a>00571
590 <a name="l00572"></a>00572 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span>
591 <a name="l00573"></a>00573 k = count >> 2u;
592 <a name="l00574"></a>00574
593 <a name="l00575"></a>00575 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span>
594 <a name="l00576"></a>00576 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span>
595 <a name="l00577"></a>00577 <span class="keywordflow">while</span>(k > 0u)
596 <a name="l00578"></a>00578 {
597 <a name="l00579"></a>00579 <span class="comment">/* Perform the multiply-accumulates */</span>
598 <a name="l00580"></a>00580 <span class="comment">/* sum += x[srcALen - srcBLen + 4] * y[3] , sum += x[srcALen - srcBLen + 3] * y[2] */</span>
599 <a name="l00581"></a>00581 sum = __SMLAD(*<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(px)++, *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(py)++, sum);
600 <a name="l00582"></a>00582 <span class="comment">/* sum += x[srcALen - srcBLen + 2] * y[1] , sum += x[srcALen - srcBLen + 1] * y[0] */</span>
601 <a name="l00583"></a>00583 sum = __SMLAD(*<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(px)++, *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(py)++, sum);
602 <a name="l00584"></a>00584
603 <a name="l00585"></a>00585 <span class="comment">/* Decrement the loop counter */</span>
604 <a name="l00586"></a>00586 k--;
605 <a name="l00587"></a>00587 }
606 <a name="l00588"></a>00588
607 <a name="l00589"></a>00589 <span class="comment">/* If the count is not a multiple of 4, compute any remaining MACs here. </span>
608 <a name="l00590"></a>00590 <span class="comment"> ** No loop unrolling is used. */</span>
609 <a name="l00591"></a>00591 k = count % 0x4u;
610 <a name="l00592"></a>00592
611 <a name="l00593"></a>00593 <span class="keywordflow">while</span>(k > 0u)
612 <a name="l00594"></a>00594 {
613 <a name="l00595"></a>00595 <span class="comment">/* Perform the multiply-accumulates */</span>
614 <a name="l00596"></a>00596 sum = __SMLAD(*px++, *py++, sum);
615 <a name="l00597"></a>00597
616 <a name="l00598"></a>00598 <span class="comment">/* Decrement the loop counter */</span>
617 <a name="l00599"></a>00599 k--;
618 <a name="l00600"></a>00600 }
619 <a name="l00601"></a>00601
620 <a name="l00602"></a>00602 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span>
621 <a name="l00603"></a>00603 *pOut = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (sum >> 15);
622 <a name="l00604"></a>00604 <span class="comment">/* Destination pointer is updated according to the address modifier, inc */</span>
623 <a name="l00605"></a>00605 pOut += inc;
624 <a name="l00606"></a>00606
625 <a name="l00607"></a>00607 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span>
626 <a name="l00608"></a>00608 px = ++pSrc1;
627 <a name="l00609"></a>00609 py = pIn2;
628 <a name="l00610"></a>00610
629 <a name="l00611"></a>00611 <span class="comment">/* Decrement the MAC count */</span>
630 <a name="l00612"></a>00612 count--;
631 <a name="l00613"></a>00613
632 <a name="l00614"></a>00614 <span class="comment">/* Decrement the loop counter */</span>
633 <a name="l00615"></a>00615 blockSize3--;
634 <a name="l00616"></a>00616 }
635 <a name="l00617"></a>00617
636 <a name="l00618"></a>00618 }
637 <a name="l00619"></a>00619
640 <!--- window showing the filter options -->
641 <div id="MSearchSelectWindow"
642 onmouseover="return searchBox.OnSearchSelectShow()"
643 onmouseout="return searchBox.OnSearchSelectHide()"
644 onkeydown="return searchBox.OnSearchSelectKey(event)">
645 <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>
647 <!-- iframe showing the search results (closed by default) -->
648 <div id="MSearchResultsWindow">
649 <iframe src="" frameborder="0"
650 name="MSearchResults" id="MSearchResults">
654 <hr class="footer"/><address class="footer"><small>Generated on Fri Jul 15 2011 13:16:16 for CMSIS DSP Software Library by 
655 <a href="http://www.doxygen.org/index.html">
656 <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.7.2 </small></address>
projects / fw / stlink / blob