| <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"> |
| <html xmlns="http://www.w3.org/1999/xhtml"> |
| <head> |
| <meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/> |
| <title>CMSIS DSP Software Library: arm_conv_fast_q15.c Source File</title> |
| <link href="tabs.css" rel="stylesheet" type="text/css"/> |
| <link href="search/search.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javaScript" src="search/search.js"></script> |
| <link href="doxygen.css" rel="stylesheet" type="text/css"/> |
| </head> |
| <body onload='searchBox.OnSelectItem(0);'> |
| <!-- Generated by Doxygen 1.7.2 --> |
| <script type="text/javascript"><!-- |
| var searchBox = new SearchBox("searchBox", "search",false,'Search'); |
| --></script> |
| <div class="navigation" id="top"> |
| <div class="tabs"> |
| <ul class="tablist"> |
| <li><a href="index.html"><span>Main Page</span></a></li> |
| <li><a href="modules.html"><span>Modules</span></a></li> |
| <li><a href="annotated.html"><span>Data Structures</span></a></li> |
| <li class="current"><a href="files.html"><span>Files</span></a></li> |
| <li><a href="examples.html"><span>Examples</span></a></li> |
| <li id="searchli"> |
| <div id="MSearchBox" class="MSearchBoxInactive"> |
| <span class="left"> |
| <img id="MSearchSelect" src="search/mag_sel.png" |
| onmouseover="return searchBox.OnSearchSelectShow()" |
| onmouseout="return searchBox.OnSearchSelectHide()" |
| alt=""/> |
| <input type="text" id="MSearchField" value="Search" accesskey="S" |
| onfocus="searchBox.OnSearchFieldFocus(true)" |
| onblur="searchBox.OnSearchFieldFocus(false)" |
| onkeyup="searchBox.OnSearchFieldChange(event)"/> |
| </span><span class="right"> |
| <a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a> |
| </span> |
| </div> |
| </li> |
| </ul> |
| </div> |
| <div class="tabs2"> |
| <ul class="tablist"> |
| <li><a href="files.html"><span>File List</span></a></li> |
| <li><a href="globals.html"><span>Globals</span></a></li> |
| </ul> |
| </div> |
| <div class="header"> |
| <div class="headertitle"> |
| <h1>arm_conv_fast_q15.c</h1> </div> |
| </div> |
| <div class="contents"> |
| <a href="arm__conv__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> |
| <a name="l00002"></a>00002 <span class="comment">* Copyright (C) 2010 ARM Limited. All rights reserved. </span> |
| <a name="l00003"></a>00003 <span class="comment">* </span> |
| <a name="l00004"></a>00004 <span class="comment">* $Date: 15. July 2011 </span> |
| <a name="l00005"></a>00005 <span class="comment">* $Revision: V1.0.10 </span> |
| <a name="l00006"></a>00006 <span class="comment">* </span> |
| <a name="l00007"></a>00007 <span class="comment">* Project: CMSIS DSP Library </span> |
| <a name="l00008"></a>00008 <span class="comment">* Title: arm_conv_fast_q15.c </span> |
| <a name="l00009"></a>00009 <span class="comment">* </span> |
| <a name="l00010"></a>00010 <span class="comment">* Description: Fast Q15 Convolution. </span> |
| <a name="l00011"></a>00011 <span class="comment">* </span> |
| <a name="l00012"></a>00012 <span class="comment">* Target Processor: Cortex-M4/Cortex-M3</span> |
| <a name="l00013"></a>00013 <span class="comment">* </span> |
| <a name="l00014"></a>00014 <span class="comment">* Version 1.0.10 2011/7/15 </span> |
| <a name="l00015"></a>00015 <span class="comment">* Big Endian support added and Merged M0 and M3/M4 Source code. </span> |
| <a name="l00016"></a>00016 <span class="comment">* </span> |
| <a name="l00017"></a>00017 <span class="comment">* Version 1.0.3 2010/11/29 </span> |
| <a name="l00018"></a>00018 <span class="comment">* Re-organized the CMSIS folders and updated documentation. </span> |
| <a name="l00019"></a>00019 <span class="comment">* </span> |
| <a name="l00020"></a>00020 <span class="comment">* Version 1.0.2 2010/11/11 </span> |
| <a name="l00021"></a>00021 <span class="comment">* Documentation updated. </span> |
| <a name="l00022"></a>00022 <span class="comment">* </span> |
| <a name="l00023"></a>00023 <span class="comment">* Version 1.0.1 2010/10/05 </span> |
| <a name="l00024"></a>00024 <span class="comment">* Production release and review comments incorporated. </span> |
| <a name="l00025"></a>00025 <span class="comment">* </span> |
| <a name="l00026"></a>00026 <span class="comment">* Version 1.0.0 2010/09/20 </span> |
| <a name="l00027"></a>00027 <span class="comment">* Production release and review comments incorporated. </span> |
| <a name="l00028"></a>00028 <span class="comment">* -------------------------------------------------------------------- */</span> |
| <a name="l00029"></a>00029 |
| <a name="l00030"></a>00030 <span class="preprocessor">#include "<a class="code" href="arm__math_8h.html">arm_math.h</a>"</span> |
| <a name="l00031"></a>00031 |
| <a name="l00066"></a><a class="code" href="group___conv.html#gad75ca978ce906e04abdf86a8d76306d4">00066</a> <span class="keywordtype">void</span> <a class="code" href="group___conv.html#gad75ca978ce906e04abdf86a8d76306d4" title="Convolution of Q15 sequences (fast version) for Cortex-M3 and Cortex-M4.">arm_conv_fast_q15</a>( |
| <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, |
| <a name="l00068"></a>00068 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>, |
| <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, |
| <a name="l00070"></a>00070 uint32_t <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>, |
| <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) |
| <a name="l00072"></a>00072 { |
| <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> |
| <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> |
| <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> |
| <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">/* Accumulator */</span> |
| <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> |
| <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> |
| <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, *pSrc2; <span class="comment">/* Intermediate pointers */</span> |
| <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 to hold state and coefficient values */</span> |
| <a name="l00081"></a>00081 uint32_t blockSize1, blockSize2, blockSize3, j, k, count, blkCnt; <span class="comment">/* loop counter */</span> |
| <a name="l00082"></a>00082 <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> |
| <a name="l00083"></a>00083 |
| <a name="l00084"></a>00084 |
| <a name="l00085"></a>00085 <span class="comment">/* The algorithm implementation is based on the lengths of the inputs. */</span> |
| <a name="l00086"></a>00086 <span class="comment">/* srcB is always made to slide across srcA. */</span> |
| <a name="l00087"></a>00087 <span class="comment">/* So srcBLen is always considered as shorter or equal to srcALen */</span> |
| <a name="l00088"></a>00088 <span class="keywordflow">if</span>(srcALen >= srcBLen) |
| <a name="l00089"></a>00089 { |
| <a name="l00090"></a>00090 <span class="comment">/* Initialization of inputA pointer */</span> |
| <a name="l00091"></a>00091 pIn1 = pSrcA; |
| <a name="l00092"></a>00092 |
| <a name="l00093"></a>00093 <span class="comment">/* Initialization of inputB pointer */</span> |
| <a name="l00094"></a>00094 pIn2 = pSrcB; |
| <a name="l00095"></a>00095 } |
| <a name="l00096"></a>00096 <span class="keywordflow">else</span> |
| <a name="l00097"></a>00097 { |
| <a name="l00098"></a>00098 <span class="comment">/* Initialization of inputA pointer */</span> |
| <a name="l00099"></a>00099 pIn1 = pSrcB; |
| <a name="l00100"></a>00100 |
| <a name="l00101"></a>00101 <span class="comment">/* Initialization of inputB pointer */</span> |
| <a name="l00102"></a>00102 pIn2 = pSrcA; |
| <a name="l00103"></a>00103 |
| <a name="l00104"></a>00104 <span class="comment">/* srcBLen is always considered as shorter or equal to srcALen */</span> |
| <a name="l00105"></a>00105 j = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>; |
| <a name="l00106"></a>00106 srcBLen = <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>; |
| <a name="l00107"></a>00107 srcALen = j; |
| <a name="l00108"></a>00108 } |
| <a name="l00109"></a>00109 |
| <a name="l00110"></a>00110 <span class="comment">/* conv(x,y) at n = x[n] * y[0] + x[n-1] * y[1] + x[n-2] * y[2] + ...+ x[n-N+1] * y[N -1] */</span> |
| <a name="l00111"></a>00111 <span class="comment">/* The function is internally </span> |
| <a name="l00112"></a>00112 <span class="comment"> * divided into three stages according to the number of multiplications that has to be </span> |
| <a name="l00113"></a>00113 <span class="comment"> * taken place between inputA samples and inputB samples. In the first stage of the </span> |
| <a name="l00114"></a>00114 <span class="comment"> * algorithm, the multiplications increase by one for every iteration. </span> |
| <a name="l00115"></a>00115 <span class="comment"> * In the second stage of the algorithm, srcBLen number of multiplications are done. </span> |
| <a name="l00116"></a>00116 <span class="comment"> * In the third stage of the algorithm, the multiplications decrease by one </span> |
| <a name="l00117"></a>00117 <span class="comment"> * for every iteration. */</span> |
| <a name="l00118"></a>00118 |
| <a name="l00119"></a>00119 <span class="comment">/* The algorithm is implemented in three stages. </span> |
| <a name="l00120"></a>00120 <span class="comment"> The loop counters of each stage is initiated here. */</span> |
| <a name="l00121"></a>00121 blockSize1 = srcBLen - 1u; |
| <a name="l00122"></a>00122 blockSize2 = srcALen - (srcBLen - 1u); |
| <a name="l00123"></a>00123 blockSize3 = blockSize1; |
| <a name="l00124"></a>00124 |
| <a name="l00125"></a>00125 <span class="comment">/* -------------------------- </span> |
| <a name="l00126"></a>00126 <span class="comment"> * Initializations of stage1 </span> |
| <a name="l00127"></a>00127 <span class="comment"> * -------------------------*/</span> |
| <a name="l00128"></a>00128 |
| <a name="l00129"></a>00129 <span class="comment">/* sum = x[0] * y[0] </span> |
| <a name="l00130"></a>00130 <span class="comment"> * sum = x[0] * y[1] + x[1] * y[0] </span> |
| <a name="l00131"></a>00131 <span class="comment"> * .... </span> |
| <a name="l00132"></a>00132 <span class="comment"> * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0] </span> |
| <a name="l00133"></a>00133 <span class="comment"> */</span> |
| <a name="l00134"></a>00134 |
| <a name="l00135"></a>00135 <span class="comment">/* In this stage the MAC operations are increased by 1 for every iteration. </span> |
| <a name="l00136"></a>00136 <span class="comment"> The count variable holds the number of MAC operations performed */</span> |
| <a name="l00137"></a>00137 count = 1u; |
| <a name="l00138"></a>00138 |
| <a name="l00139"></a>00139 <span class="comment">/* Working pointer of inputA */</span> |
| <a name="l00140"></a>00140 px = pIn1; |
| <a name="l00141"></a>00141 |
| <a name="l00142"></a>00142 <span class="comment">/* Working pointer of inputB */</span> |
| <a name="l00143"></a>00143 py = pIn2; |
| <a name="l00144"></a>00144 |
| <a name="l00145"></a>00145 |
| <a name="l00146"></a>00146 <span class="comment">/* ------------------------ </span> |
| <a name="l00147"></a>00147 <span class="comment"> * Stage1 process </span> |
| <a name="l00148"></a>00148 <span class="comment"> * ----------------------*/</span> |
| <a name="l00149"></a>00149 |
| <a name="l00150"></a>00150 <span class="comment">/* For loop unrolling by 4, this stage is divided into two. */</span> |
| <a name="l00151"></a>00151 <span class="comment">/* First part of this stage computes the MAC operations less than 4 */</span> |
| <a name="l00152"></a>00152 <span class="comment">/* Second part of this stage computes the MAC operations greater than or equal to 4 */</span> |
| <a name="l00153"></a>00153 |
| <a name="l00154"></a>00154 <span class="comment">/* The first part of the stage starts here */</span> |
| <a name="l00155"></a>00155 <span class="keywordflow">while</span>((count < 4u) && (blockSize1 > 0u)) |
| <a name="l00156"></a>00156 { |
| <a name="l00157"></a>00157 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00158"></a>00158 sum = 0; |
| <a name="l00159"></a>00159 |
| <a name="l00160"></a>00160 <span class="comment">/* Loop over number of MAC operations between </span> |
| <a name="l00161"></a>00161 <span class="comment"> * inputA samples and inputB samples */</span> |
| <a name="l00162"></a>00162 k = count; |
| <a name="l00163"></a>00163 |
| <a name="l00164"></a>00164 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00165"></a>00165 { |
| <a name="l00166"></a>00166 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00167"></a>00167 sum = __SMLAD(*px++, *py--, sum); |
| <a name="l00168"></a>00168 |
| <a name="l00169"></a>00169 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00170"></a>00170 k--; |
| <a name="l00171"></a>00171 } |
| <a name="l00172"></a>00172 |
| <a name="l00173"></a>00173 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00174"></a>00174 *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); |
| <a name="l00175"></a>00175 |
| <a name="l00176"></a>00176 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00177"></a>00177 py = pIn2 + count; |
| <a name="l00178"></a>00178 px = pIn1; |
| <a name="l00179"></a>00179 |
| <a name="l00180"></a>00180 <span class="comment">/* Increment the MAC count */</span> |
| <a name="l00181"></a>00181 count++; |
| <a name="l00182"></a>00182 |
| <a name="l00183"></a>00183 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00184"></a>00184 blockSize1--; |
| <a name="l00185"></a>00185 } |
| <a name="l00186"></a>00186 |
| <a name="l00187"></a>00187 <span class="comment">/* The second part of the stage starts here */</span> |
| <a name="l00188"></a>00188 <span class="comment">/* The internal loop, over count, is unrolled by 4 */</span> |
| <a name="l00189"></a>00189 <span class="comment">/* To, read the last two inputB samples using SIMD: </span> |
| <a name="l00190"></a>00190 <span class="comment"> * y[srcBLen] and y[srcBLen-1] coefficients, py is decremented by 1 */</span> |
| <a name="l00191"></a>00191 py = py - 1; |
| <a name="l00192"></a>00192 |
| <a name="l00193"></a>00193 <span class="keywordflow">while</span>(blockSize1 > 0u) |
| <a name="l00194"></a>00194 { |
| <a name="l00195"></a>00195 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00196"></a>00196 sum = 0; |
| <a name="l00197"></a>00197 |
| <a name="l00198"></a>00198 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span> |
| <a name="l00199"></a>00199 k = count >> 2u; |
| <a name="l00200"></a>00200 |
| <a name="l00201"></a>00201 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span> |
| <a name="l00202"></a>00202 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span> |
| <a name="l00203"></a>00203 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00204"></a>00204 { |
| <a name="l00205"></a>00205 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00206"></a>00206 <span class="comment">/* x[0], x[1] are multiplied with y[srcBLen - 1], y[srcBLen - 2] respectively */</span> |
| <a name="l00207"></a>00207 sum = __SMLADX(*<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); |
| <a name="l00208"></a>00208 <span class="comment">/* x[2], x[3] are multiplied with y[srcBLen - 3], y[srcBLen - 4] respectively */</span> |
| <a name="l00209"></a>00209 sum = __SMLADX(*<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); |
| <a name="l00210"></a>00210 |
| <a name="l00211"></a>00211 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00212"></a>00212 k--; |
| <a name="l00213"></a>00213 } |
| <a name="l00214"></a>00214 |
| <a name="l00215"></a>00215 <span class="comment">/* For the next MAC operations, the pointer py is used without SIMD </span> |
| <a name="l00216"></a>00216 <span class="comment"> * So, py is incremented by 1 */</span> |
| <a name="l00217"></a>00217 py = py + 1u; |
| <a name="l00218"></a>00218 |
| <a name="l00219"></a>00219 <span class="comment">/* If the count is not a multiple of 4, compute any remaining MACs here. </span> |
| <a name="l00220"></a>00220 <span class="comment"> ** No loop unrolling is used. */</span> |
| <a name="l00221"></a>00221 k = count % 0x4u; |
| <a name="l00222"></a>00222 |
| <a name="l00223"></a>00223 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00224"></a>00224 { |
| <a name="l00225"></a>00225 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00226"></a>00226 sum = __SMLAD(*px++, *py--, sum); |
| <a name="l00227"></a>00227 |
| <a name="l00228"></a>00228 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00229"></a>00229 k--; |
| <a name="l00230"></a>00230 } |
| <a name="l00231"></a>00231 |
| <a name="l00232"></a>00232 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00233"></a>00233 *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); |
| <a name="l00234"></a>00234 |
| <a name="l00235"></a>00235 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00236"></a>00236 py = pIn2 + (count - 1u); |
| <a name="l00237"></a>00237 px = pIn1; |
| <a name="l00238"></a>00238 |
| <a name="l00239"></a>00239 <span class="comment">/* Increment the MAC count */</span> |
| <a name="l00240"></a>00240 count++; |
| <a name="l00241"></a>00241 |
| <a name="l00242"></a>00242 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00243"></a>00243 blockSize1--; |
| <a name="l00244"></a>00244 } |
| <a name="l00245"></a>00245 |
| <a name="l00246"></a>00246 <span class="comment">/* -------------------------- </span> |
| <a name="l00247"></a>00247 <span class="comment"> * Initializations of stage2 </span> |
| <a name="l00248"></a>00248 <span class="comment"> * ------------------------*/</span> |
| <a name="l00249"></a>00249 |
| <a name="l00250"></a>00250 <span class="comment">/* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0] </span> |
| <a name="l00251"></a>00251 <span class="comment"> * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0] </span> |
| <a name="l00252"></a>00252 <span class="comment"> * .... </span> |
| <a name="l00253"></a>00253 <span class="comment"> * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0] </span> |
| <a name="l00254"></a>00254 <span class="comment"> */</span> |
| <a name="l00255"></a>00255 |
| <a name="l00256"></a>00256 <span class="comment">/* Working pointer of inputA */</span> |
| <a name="l00257"></a>00257 px = pIn1; |
| <a name="l00258"></a>00258 |
| <a name="l00259"></a>00259 <span class="comment">/* Working pointer of inputB */</span> |
| <a name="l00260"></a>00260 pSrc2 = pIn2 + (srcBLen - 1u); |
| <a name="l00261"></a>00261 py = pSrc2; |
| <a name="l00262"></a>00262 |
| <a name="l00263"></a>00263 <span class="comment">/* Initialize inputB pointer of type q31 */</span> |
| <a name="l00264"></a>00264 pb = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (py - 1u); |
| <a name="l00265"></a>00265 |
| <a name="l00266"></a>00266 <span class="comment">/* count is the index by which the pointer pIn1 to be incremented */</span> |
| <a name="l00267"></a>00267 count = 1u; |
| <a name="l00268"></a>00268 |
| <a name="l00269"></a>00269 |
| <a name="l00270"></a>00270 <span class="comment">/* -------------------- </span> |
| <a name="l00271"></a>00271 <span class="comment"> * Stage2 process </span> |
| <a name="l00272"></a>00272 <span class="comment"> * -------------------*/</span> |
| <a name="l00273"></a>00273 |
| <a name="l00274"></a>00274 <span class="comment">/* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. </span> |
| <a name="l00275"></a>00275 <span class="comment"> * So, to loop unroll over blockSize2, </span> |
| <a name="l00276"></a>00276 <span class="comment"> * srcBLen should be greater than or equal to 4 */</span> |
| <a name="l00277"></a>00277 <span class="keywordflow">if</span>(srcBLen >= 4u) |
| <a name="l00278"></a>00278 { |
| <a name="l00279"></a>00279 <span class="comment">/* Loop unroll over blockSize2, by 4 */</span> |
| <a name="l00280"></a>00280 blkCnt = blockSize2 >> 2u; |
| <a name="l00281"></a>00281 |
| <a name="l00282"></a>00282 <span class="keywordflow">while</span>(blkCnt > 0u) |
| <a name="l00283"></a>00283 { |
| <a name="l00284"></a>00284 <span class="comment">/* Set all accumulators to zero */</span> |
| <a name="l00285"></a>00285 acc0 = 0; |
| <a name="l00286"></a>00286 acc1 = 0; |
| <a name="l00287"></a>00287 acc2 = 0; |
| <a name="l00288"></a>00288 acc3 = 0; |
| <a name="l00289"></a>00289 |
| <a name="l00290"></a>00290 |
| <a name="l00291"></a>00291 <span class="comment">/* read x[0], x[1] samples */</span> |
| <a name="l00292"></a>00292 x0 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00293"></a>00293 <span class="comment">/* read x[1], x[2] samples */</span> |
| <a name="l00294"></a>00294 x1 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00295"></a>00295 |
| <a name="l00296"></a>00296 |
| <a name="l00297"></a>00297 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span> |
| <a name="l00298"></a>00298 k = srcBLen >> 2u; |
| <a name="l00299"></a>00299 |
| <a name="l00300"></a>00300 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span> |
| <a name="l00301"></a>00301 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span> |
| <a name="l00302"></a>00302 <span class="keywordflow">do</span> |
| <a name="l00303"></a>00303 { |
| <a name="l00304"></a>00304 <span class="comment">/* Read the last two inputB samples using SIMD: </span> |
| <a name="l00305"></a>00305 <span class="comment"> * y[srcBLen - 1] and y[srcBLen - 2] */</span> |
| <a name="l00306"></a>00306 c0 = *(pb--); |
| <a name="l00307"></a>00307 |
| <a name="l00308"></a>00308 <span class="comment">/* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */</span> |
| <a name="l00309"></a>00309 acc0 = __SMLADX(x0, c0, acc0); |
| <a name="l00310"></a>00310 |
| <a name="l00311"></a>00311 <span class="comment">/* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */</span> |
| <a name="l00312"></a>00312 acc1 = __SMLADX(x1, c0, acc1); |
| <a name="l00313"></a>00313 |
| <a name="l00314"></a>00314 <span class="comment">/* Read x[2], x[3] */</span> |
| <a name="l00315"></a>00315 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00316"></a>00316 |
| <a name="l00317"></a>00317 <span class="comment">/* Read x[3], x[4] */</span> |
| <a name="l00318"></a>00318 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00319"></a>00319 |
| <a name="l00320"></a>00320 <span class="comment">/* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */</span> |
| <a name="l00321"></a>00321 acc2 = __SMLADX(x2, c0, acc2); |
| <a name="l00322"></a>00322 |
| <a name="l00323"></a>00323 <span class="comment">/* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */</span> |
| <a name="l00324"></a>00324 acc3 = __SMLADX(x3, c0, acc3); |
| <a name="l00325"></a>00325 |
| <a name="l00326"></a>00326 <span class="comment">/* Read y[srcBLen - 3] and y[srcBLen - 4] */</span> |
| <a name="l00327"></a>00327 c0 = *(pb--); |
| <a name="l00328"></a>00328 |
| <a name="l00329"></a>00329 <span class="comment">/* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */</span> |
| <a name="l00330"></a>00330 acc0 = __SMLADX(x2, c0, acc0); |
| <a name="l00331"></a>00331 |
| <a name="l00332"></a>00332 <span class="comment">/* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */</span> |
| <a name="l00333"></a>00333 acc1 = __SMLADX(x3, c0, acc1); |
| <a name="l00334"></a>00334 |
| <a name="l00335"></a>00335 <span class="comment">/* Read x[4], x[5] */</span> |
| <a name="l00336"></a>00336 x0 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00337"></a>00337 |
| <a name="l00338"></a>00338 <span class="comment">/* Read x[5], x[6] */</span> |
| <a name="l00339"></a>00339 x1 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (px++); |
| <a name="l00340"></a>00340 |
| <a name="l00341"></a>00341 <span class="comment">/* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */</span> |
| <a name="l00342"></a>00342 acc2 = __SMLADX(x0, c0, acc2); |
| <a name="l00343"></a>00343 |
| <a name="l00344"></a>00344 <span class="comment">/* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */</span> |
| <a name="l00345"></a>00345 acc3 = __SMLADX(x1, c0, acc3); |
| <a name="l00346"></a>00346 |
| <a name="l00347"></a>00347 } <span class="keywordflow">while</span>(--k); |
| <a name="l00348"></a>00348 |
| <a name="l00349"></a>00349 <span class="comment">/* For the next MAC operations, SIMD is not used </span> |
| <a name="l00350"></a>00350 <span class="comment"> * So, the 16 bit pointer if inputB, py is updated */</span> |
| <a name="l00351"></a>00351 py = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a> *) pb; |
| <a name="l00352"></a>00352 py = py + 1; |
| <a name="l00353"></a>00353 |
| <a name="l00354"></a>00354 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span> |
| <a name="l00355"></a>00355 <span class="comment"> ** No loop unrolling is used. */</span> |
| <a name="l00356"></a>00356 k = srcBLen % 0x4u; |
| <a name="l00357"></a>00357 |
| <a name="l00358"></a>00358 <span class="keywordflow">if</span>(k == 1u) |
| <a name="l00359"></a>00359 { |
| <a name="l00360"></a>00360 <span class="comment">/* Read y[srcBLen - 5] */</span> |
| <a name="l00361"></a>00361 c0 = *(py); |
| <a name="l00362"></a>00362 <span class="preprocessor">#ifdef ARM_MATH_BIG_ENDIAN</span> |
| <a name="l00363"></a>00363 <span class="preprocessor"></span> |
| <a name="l00364"></a>00364 <span class="comment">// c0 = unallign_rev(p, c0); </span> |
| <a name="l00365"></a>00365 c0 = c0 << 16; |
| <a name="l00366"></a>00366 <span class="preprocessor">#endif </span><span class="comment">/* #ifdef ARM_MATH_BIG_ENDIAN */</span> |
| <a name="l00367"></a>00367 |
| <a name="l00368"></a>00368 <span class="comment">/* Read x[7] */</span> |
| <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++; |
| <a name="l00370"></a>00370 |
| <a name="l00371"></a>00371 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00372"></a>00372 acc0 = __SMLAD(x0, c0, acc0); |
| <a name="l00373"></a>00373 acc1 = __SMLAD(x1, c0, acc1); |
| <a name="l00374"></a>00374 acc2 = __SMLADX(x1, c0, acc2); |
| <a name="l00375"></a>00375 acc3 = __SMLADX(x3, c0, acc3); |
| <a name="l00376"></a>00376 } |
| <a name="l00377"></a>00377 |
| <a name="l00378"></a>00378 <span class="keywordflow">if</span>(k == 2u) |
| <a name="l00379"></a>00379 { |
| <a name="l00380"></a>00380 <span class="comment">/* Read y[srcBLen - 5], y[srcBLen - 6] */</span> |
| <a name="l00381"></a>00381 c0 = *(pb); |
| <a name="l00382"></a>00382 |
| <a name="l00383"></a>00383 <span class="comment">/* Read x[7], x[8] */</span> |
| <a name="l00384"></a>00384 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++; |
| <a name="l00385"></a>00385 |
| <a name="l00386"></a>00386 <span class="comment">/* Read x[9] */</span> |
| <a name="l00387"></a>00387 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++; |
| <a name="l00388"></a>00388 |
| <a name="l00389"></a>00389 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00390"></a>00390 acc0 = __SMLADX(x0, c0, acc0); |
| <a name="l00391"></a>00391 acc1 = __SMLADX(x1, c0, acc1); |
| <a name="l00392"></a>00392 acc2 = __SMLADX(x3, c0, acc2); |
| <a name="l00393"></a>00393 acc3 = __SMLADX(x2, c0, acc3); |
| <a name="l00394"></a>00394 } |
| <a name="l00395"></a>00395 |
| <a name="l00396"></a>00396 <span class="keywordflow">if</span>(k == 3u) |
| <a name="l00397"></a>00397 { |
| <a name="l00398"></a>00398 <span class="comment">/* Read y[srcBLen - 5], y[srcBLen - 6] */</span> |
| <a name="l00399"></a>00399 c0 = *pb--; |
| <a name="l00400"></a>00400 |
| <a name="l00401"></a>00401 <span class="comment">/* Read x[7], x[8] */</span> |
| <a name="l00402"></a>00402 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++; |
| <a name="l00403"></a>00403 |
| <a name="l00404"></a>00404 <span class="comment">/* Read x[9] */</span> |
| <a name="l00405"></a>00405 x2 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++; |
| <a name="l00406"></a>00406 |
| <a name="l00407"></a>00407 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00408"></a>00408 acc0 = __SMLADX(x0, c0, acc0); |
| <a name="l00409"></a>00409 acc1 = __SMLADX(x1, c0, acc1); |
| <a name="l00410"></a>00410 acc2 = __SMLADX(x3, c0, acc2); |
| <a name="l00411"></a>00411 acc3 = __SMLADX(x2, c0, acc3); |
| <a name="l00412"></a>00412 |
| <a name="l00413"></a>00413 <span class="comment">/* Read y[srcBLen - 7] */</span> |
| <a name="l00414"></a>00414 <span class="preprocessor">#ifdef ARM_MATH_BIG_ENDIAN</span> |
| <a name="l00415"></a>00415 <span class="preprocessor"></span> |
| <a name="l00416"></a>00416 c0 = (*pb); |
| <a name="l00417"></a>00417 <span class="comment">// c0 = (c0 & 0x0000FFFF)<<16; </span> |
| <a name="l00418"></a>00418 c0 = (c0) << 16; |
| <a name="l00419"></a>00419 |
| <a name="l00420"></a>00420 <span class="preprocessor">#else</span> |
| <a name="l00421"></a>00421 <span class="preprocessor"></span> |
| <a name="l00422"></a>00422 c0 = (<a class="code" href="arm__math_8h.html#ab5a8fb21a5b3b983d5f54f31614052ea" title="16-bit fractional data type in 1.15 format.">q15_t</a>) (*pb >> 16); |
| <a name="l00423"></a>00423 |
| <a name="l00424"></a>00424 <span class="preprocessor">#endif </span><span class="comment">/* #ifdef ARM_MATH_BIG_ENDIAN */</span> |
| <a name="l00425"></a>00425 |
| <a name="l00426"></a>00426 <span class="comment">/* Read x[10] */</span> |
| <a name="l00427"></a>00427 x3 = *(<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) px++; |
| <a name="l00428"></a>00428 |
| <a name="l00429"></a>00429 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00430"></a>00430 acc0 = __SMLADX(x1, c0, acc0); |
| <a name="l00431"></a>00431 acc1 = __SMLAD(x2, c0, acc1); |
| <a name="l00432"></a>00432 acc2 = __SMLADX(x2, c0, acc2); |
| <a name="l00433"></a>00433 acc3 = __SMLADX(x3, c0, acc3); |
| <a name="l00434"></a>00434 } |
| <a name="l00435"></a>00435 |
| <a name="l00436"></a>00436 <span class="comment">/* Store the results in the accumulators in the destination buffer. */</span> |
| <a name="l00437"></a>00437 <span class="preprocessor">#ifndef ARM_MATH_BIG_ENDIAN</span> |
| <a name="l00438"></a>00438 <span class="preprocessor"></span> |
| <a name="l00439"></a>00439 *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(pOut)++ = __PKHBT((acc0 >> 15), (acc1 >> 15), 16); |
| <a name="l00440"></a>00440 *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(pOut)++ = __PKHBT((acc2 >> 15), (acc3 >> 15), 16); |
| <a name="l00441"></a>00441 |
| <a name="l00442"></a>00442 <span class="preprocessor">#else</span> |
| <a name="l00443"></a>00443 <span class="preprocessor"></span> |
| <a name="l00444"></a>00444 *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(pOut)++ = __PKHBT((acc1 >> 15), (acc0 >> 15), 16); |
| <a name="l00445"></a>00445 *<a class="code" href="arm__math_8h.html#a9de2e0a5785be82866bcb96012282248" title="definition to read/write two 16 bit values.">__SIMD32</a>(pOut)++ = __PKHBT((acc3 >> 15), (acc2 >> 15), 16); |
| <a name="l00446"></a>00446 |
| <a name="l00447"></a>00447 <span class="preprocessor">#endif </span><span class="comment">/* #ifndef ARM_MATH_BIG_ENDIAN */</span> |
| <a name="l00448"></a>00448 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00449"></a>00449 px = pIn1 + (count * 4u); |
| <a name="l00450"></a>00450 py = pSrc2; |
| <a name="l00451"></a>00451 pb = (<a class="code" href="arm__math_8h.html#adc89a3547f5324b7b3b95adec3806bc0" title="32-bit fractional data type in 1.31 format.">q31_t</a> *) (py - 1); |
| <a name="l00452"></a>00452 |
| <a name="l00453"></a>00453 <span class="comment">/* Increment the pointer pIn1 index, count by 1 */</span> |
| <a name="l00454"></a>00454 count++; |
| <a name="l00455"></a>00455 |
| <a name="l00456"></a>00456 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00457"></a>00457 blkCnt--; |
| <a name="l00458"></a>00458 } |
| <a name="l00459"></a>00459 |
| <a name="l00460"></a>00460 <span class="comment">/* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. </span> |
| <a name="l00461"></a>00461 <span class="comment"> ** No loop unrolling is used. */</span> |
| <a name="l00462"></a>00462 blkCnt = blockSize2 % 0x4u; |
| <a name="l00463"></a>00463 |
| <a name="l00464"></a>00464 <span class="keywordflow">while</span>(blkCnt > 0u) |
| <a name="l00465"></a>00465 { |
| <a name="l00466"></a>00466 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00467"></a>00467 sum = 0; |
| <a name="l00468"></a>00468 |
| <a name="l00469"></a>00469 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span> |
| <a name="l00470"></a>00470 k = srcBLen >> 2u; |
| <a name="l00471"></a>00471 |
| <a name="l00472"></a>00472 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span> |
| <a name="l00473"></a>00473 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span> |
| <a name="l00474"></a>00474 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00475"></a>00475 { |
| <a name="l00476"></a>00476 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00477"></a>00477 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--); |
| <a name="l00478"></a>00478 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--); |
| <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--); |
| <a name="l00480"></a>00480 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--); |
| <a name="l00481"></a>00481 |
| <a name="l00482"></a>00482 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00483"></a>00483 k--; |
| <a name="l00484"></a>00484 } |
| <a name="l00485"></a>00485 |
| <a name="l00486"></a>00486 <span class="comment">/* If the srcBLen is not a multiple of 4, compute any remaining MACs here. </span> |
| <a name="l00487"></a>00487 <span class="comment"> ** No loop unrolling is used. */</span> |
| <a name="l00488"></a>00488 k = srcBLen % 0x4u; |
| <a name="l00489"></a>00489 |
| <a name="l00490"></a>00490 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00491"></a>00491 { |
| <a name="l00492"></a>00492 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00493"></a>00493 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--); |
| <a name="l00494"></a>00494 |
| <a name="l00495"></a>00495 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00496"></a>00496 k--; |
| <a name="l00497"></a>00497 } |
| <a name="l00498"></a>00498 |
| <a name="l00499"></a>00499 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00500"></a>00500 *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); |
| <a name="l00501"></a>00501 |
| <a name="l00502"></a>00502 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00503"></a>00503 px = pIn1 + count; |
| <a name="l00504"></a>00504 py = pSrc2; |
| <a name="l00505"></a>00505 |
| <a name="l00506"></a>00506 <span class="comment">/* Increment the pointer pIn1 index, count by 1 */</span> |
| <a name="l00507"></a>00507 count++; |
| <a name="l00508"></a>00508 |
| <a name="l00509"></a>00509 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00510"></a>00510 blkCnt--; |
| <a name="l00511"></a>00511 } |
| <a name="l00512"></a>00512 } |
| <a name="l00513"></a>00513 <span class="keywordflow">else</span> |
| <a name="l00514"></a>00514 { |
| <a name="l00515"></a>00515 <span class="comment">/* If the srcBLen is not a multiple of 4, </span> |
| <a name="l00516"></a>00516 <span class="comment"> * the blockSize2 loop cannot be unrolled by 4 */</span> |
| <a name="l00517"></a>00517 blkCnt = blockSize2; |
| <a name="l00518"></a>00518 |
| <a name="l00519"></a>00519 <span class="keywordflow">while</span>(blkCnt > 0u) |
| <a name="l00520"></a>00520 { |
| <a name="l00521"></a>00521 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00522"></a>00522 sum = 0; |
| <a name="l00523"></a>00523 |
| <a name="l00524"></a>00524 <span class="comment">/* srcBLen number of MACS should be performed */</span> |
| <a name="l00525"></a>00525 k = <a class="code" href="arm__convolution__example__f32_8c.html#aea71286f498978c5ed3775609b974fc8">srcBLen</a>; |
| <a name="l00526"></a>00526 |
| <a name="l00527"></a>00527 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00528"></a>00528 { |
| <a name="l00529"></a>00529 <span class="comment">/* Perform the multiply-accumulate */</span> |
| <a name="l00530"></a>00530 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--); |
| <a name="l00531"></a>00531 |
| <a name="l00532"></a>00532 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00533"></a>00533 k--; |
| <a name="l00534"></a>00534 } |
| <a name="l00535"></a>00535 |
| <a name="l00536"></a>00536 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00537"></a>00537 *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); |
| <a name="l00538"></a>00538 |
| <a name="l00539"></a>00539 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00540"></a>00540 px = pIn1 + count; |
| <a name="l00541"></a>00541 py = pSrc2; |
| <a name="l00542"></a>00542 |
| <a name="l00543"></a>00543 <span class="comment">/* Increment the MAC count */</span> |
| <a name="l00544"></a>00544 count++; |
| <a name="l00545"></a>00545 |
| <a name="l00546"></a>00546 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00547"></a>00547 blkCnt--; |
| <a name="l00548"></a>00548 } |
| <a name="l00549"></a>00549 } |
| <a name="l00550"></a>00550 |
| <a name="l00551"></a>00551 |
| <a name="l00552"></a>00552 <span class="comment">/* -------------------------- </span> |
| <a name="l00553"></a>00553 <span class="comment"> * Initializations of stage3 </span> |
| <a name="l00554"></a>00554 <span class="comment"> * -------------------------*/</span> |
| <a name="l00555"></a>00555 |
| <a name="l00556"></a>00556 <span class="comment">/* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1] </span> |
| <a name="l00557"></a>00557 <span class="comment"> * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2] </span> |
| <a name="l00558"></a>00558 <span class="comment"> * .... </span> |
| <a name="l00559"></a>00559 <span class="comment"> * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2] </span> |
| <a name="l00560"></a>00560 <span class="comment"> * sum += x[srcALen-1] * y[srcBLen-1] </span> |
| <a name="l00561"></a>00561 <span class="comment"> */</span> |
| <a name="l00562"></a>00562 |
| <a name="l00563"></a>00563 <span class="comment">/* In this stage the MAC operations are decreased by 1 for every iteration. </span> |
| <a name="l00564"></a>00564 <span class="comment"> The blockSize3 variable holds the number of MAC operations performed */</span> |
| <a name="l00565"></a>00565 |
| <a name="l00566"></a>00566 <span class="comment">/* Working pointer of inputA */</span> |
| <a name="l00567"></a>00567 pSrc1 = (pIn1 + <a class="code" href="arm__convolution__example__f32_8c.html#ace48ed566e2cd6a680f0681192e6af28">srcALen</a>) - (srcBLen - 1u); |
| <a name="l00568"></a>00568 px = pSrc1; |
| <a name="l00569"></a>00569 |
| <a name="l00570"></a>00570 <span class="comment">/* Working pointer of inputB */</span> |
| <a name="l00571"></a>00571 pSrc2 = pIn2 + (srcBLen - 1u); |
| <a name="l00572"></a>00572 pIn2 = pSrc2 - 1u; |
| <a name="l00573"></a>00573 py = pIn2; |
| <a name="l00574"></a>00574 |
| <a name="l00575"></a>00575 <span class="comment">/* ------------------- </span> |
| <a name="l00576"></a>00576 <span class="comment"> * Stage3 process </span> |
| <a name="l00577"></a>00577 <span class="comment"> * ------------------*/</span> |
| <a name="l00578"></a>00578 |
| <a name="l00579"></a>00579 <span class="comment">/* For loop unrolling by 4, this stage is divided into two. */</span> |
| <a name="l00580"></a>00580 <span class="comment">/* First part of this stage computes the MAC operations greater than 4 */</span> |
| <a name="l00581"></a>00581 <span class="comment">/* Second part of this stage computes the MAC operations less than or equal to 4 */</span> |
| <a name="l00582"></a>00582 |
| <a name="l00583"></a>00583 <span class="comment">/* The first part of the stage starts here */</span> |
| <a name="l00584"></a>00584 j = blockSize3 >> 2u; |
| <a name="l00585"></a>00585 |
| <a name="l00586"></a>00586 <span class="keywordflow">while</span>((j > 0u) && (blockSize3 > 0u)) |
| <a name="l00587"></a>00587 { |
| <a name="l00588"></a>00588 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00589"></a>00589 sum = 0; |
| <a name="l00590"></a>00590 |
| <a name="l00591"></a>00591 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span> |
| <a name="l00592"></a>00592 k = blockSize3 >> 2u; |
| <a name="l00593"></a>00593 |
| <a name="l00594"></a>00594 <span class="comment">/* First part of the processing with loop unrolling. Compute 4 MACs at a time. </span> |
| <a name="l00595"></a>00595 <span class="comment"> ** a second loop below computes MACs for the remaining 1 to 3 samples. */</span> |
| <a name="l00596"></a>00596 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00597"></a>00597 { |
| <a name="l00598"></a>00598 <span class="comment">/* x[srcALen - srcBLen + 1], x[srcALen - srcBLen + 2] are multiplied </span> |
| <a name="l00599"></a>00599 <span class="comment"> * with y[srcBLen - 1], y[srcBLen - 2] respectively */</span> |
| <a name="l00600"></a>00600 sum = __SMLADX(*<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); |
| <a name="l00601"></a>00601 <span class="comment">/* x[srcALen - srcBLen + 3], x[srcALen - srcBLen + 4] are multiplied </span> |
| <a name="l00602"></a>00602 <span class="comment"> * with y[srcBLen - 3], y[srcBLen - 4] respectively */</span> |
| <a name="l00603"></a>00603 sum = __SMLADX(*<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); |
| <a name="l00604"></a>00604 |
| <a name="l00605"></a>00605 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00606"></a>00606 k--; |
| <a name="l00607"></a>00607 } |
| <a name="l00608"></a>00608 |
| <a name="l00609"></a>00609 <span class="comment">/* For the next MAC operations, the pointer py is used without SIMD </span> |
| <a name="l00610"></a>00610 <span class="comment"> * So, py is incremented by 1 */</span> |
| <a name="l00611"></a>00611 py = py + 1u; |
| <a name="l00612"></a>00612 |
| <a name="l00613"></a>00613 <span class="comment">/* If the blockSize3 is not a multiple of 4, compute any remaining MACs here. </span> |
| <a name="l00614"></a>00614 <span class="comment"> ** No loop unrolling is used. */</span> |
| <a name="l00615"></a>00615 k = blockSize3 % 0x4u; |
| <a name="l00616"></a>00616 |
| <a name="l00617"></a>00617 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00618"></a>00618 { |
| <a name="l00619"></a>00619 <span class="comment">/* sum += x[srcALen - srcBLen + 5] * y[srcBLen - 5] */</span> |
| <a name="l00620"></a>00620 sum = __SMLAD(*px++, *py--, sum); |
| <a name="l00621"></a>00621 |
| <a name="l00622"></a>00622 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00623"></a>00623 k--; |
| <a name="l00624"></a>00624 } |
| <a name="l00625"></a>00625 |
| <a name="l00626"></a>00626 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00627"></a>00627 *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); |
| <a name="l00628"></a>00628 |
| <a name="l00629"></a>00629 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00630"></a>00630 px = ++pSrc1; |
| <a name="l00631"></a>00631 py = pIn2; |
| <a name="l00632"></a>00632 |
| <a name="l00633"></a>00633 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00634"></a>00634 blockSize3--; |
| <a name="l00635"></a>00635 |
| <a name="l00636"></a>00636 j--; |
| <a name="l00637"></a>00637 } |
| <a name="l00638"></a>00638 |
| <a name="l00639"></a>00639 <span class="comment">/* The second part of the stage starts here */</span> |
| <a name="l00640"></a>00640 <span class="comment">/* SIMD is not used for the next MAC operations, </span> |
| <a name="l00641"></a>00641 <span class="comment"> * so pointer py is updated to read only one sample at a time */</span> |
| <a name="l00642"></a>00642 py = py + 1u; |
| <a name="l00643"></a>00643 |
| <a name="l00644"></a>00644 <span class="keywordflow">while</span>(blockSize3 > 0u) |
| <a name="l00645"></a>00645 { |
| <a name="l00646"></a>00646 <span class="comment">/* Accumulator is made zero for every iteration */</span> |
| <a name="l00647"></a>00647 sum = 0; |
| <a name="l00648"></a>00648 |
| <a name="l00649"></a>00649 <span class="comment">/* Apply loop unrolling and compute 4 MACs simultaneously. */</span> |
| <a name="l00650"></a>00650 k = blockSize3; |
| <a name="l00651"></a>00651 |
| <a name="l00652"></a>00652 <span class="keywordflow">while</span>(k > 0u) |
| <a name="l00653"></a>00653 { |
| <a name="l00654"></a>00654 <span class="comment">/* Perform the multiply-accumulates */</span> |
| <a name="l00655"></a>00655 <span class="comment">/* sum += x[srcALen-1] * y[srcBLen-1] */</span> |
| <a name="l00656"></a>00656 sum = __SMLAD(*px++, *py--, sum); |
| <a name="l00657"></a>00657 |
| <a name="l00658"></a>00658 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00659"></a>00659 k--; |
| <a name="l00660"></a>00660 } |
| <a name="l00661"></a>00661 |
| <a name="l00662"></a>00662 <span class="comment">/* Store the result in the accumulator in the destination buffer. */</span> |
| <a name="l00663"></a>00663 *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); |
| <a name="l00664"></a>00664 |
| <a name="l00665"></a>00665 <span class="comment">/* Update the inputA and inputB pointers for next MAC calculation */</span> |
| <a name="l00666"></a>00666 px = ++pSrc1; |
| <a name="l00667"></a>00667 py = pSrc2; |
| <a name="l00668"></a>00668 |
| <a name="l00669"></a>00669 <span class="comment">/* Decrement the loop counter */</span> |
| <a name="l00670"></a>00670 blockSize3--; |
| <a name="l00671"></a>00671 } |
| <a name="l00672"></a>00672 |
| <a name="l00673"></a>00673 } |
| <a name="l00674"></a>00674 |
| </pre></div></div> |
| </div> |
| <!--- window showing the filter options --> |
| <div id="MSearchSelectWindow" |
| onmouseover="return searchBox.OnSearchSelectShow()" |
| onmouseout="return searchBox.OnSearchSelectHide()" |
| onkeydown="return searchBox.OnSearchSelectKey(event)"> |
| <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> |
| |
| <!-- iframe showing the search results (closed by default) --> |
| <div id="MSearchResultsWindow"> |
| <iframe src="" frameborder="0" |
| name="MSearchResults" id="MSearchResults"> |
| </iframe> |
| </div> |
| |
| <hr class="footer"/><address class="footer"><small>Generated on Fri Jul 15 2011 13:16:16 for CMSIS DSP Software Library by  |
| <a href="http://www.doxygen.org/index.html"> |
| <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.7.2 </small></address> |
| </body> |
| </html> |
