| <!-- HTML header for doxygen 1.8.9.1--> |
| <!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"/> |
| <meta http-equiv="X-UA-Compatible" content="IE=9"/> |
| <meta name="generator" content="Doxygen 1.8.6"/> |
| <meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines --> |
| <title>Compute Library: src/core/CL/cl_kernels/gemm.cl Source File</title> |
| <link href="tabs.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javascript" src="jquery.js"></script> |
| <script type="text/javascript" src="dynsections.js"></script> |
| <link href="navtree.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javascript" src="resize.js"></script> |
| <script type="text/javascript" src="navtree.js"></script> |
| <script type="text/javascript"> |
| $(document).ready(initResizable); |
| $(window).load(resizeHeight); |
| </script> |
| <link href="search/search.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javascript" src="search/search.js"></script> |
| <script type="text/javascript"> |
| $(document).ready(function() { searchBox.OnSelectItem(0); }); |
| </script> |
| <script type="text/x-mathjax-config"> |
| MathJax.Hub.Config({ |
| extensions: ["tex2jax.js"], |
| jax: ["input/TeX","output/HTML-CSS"], |
| }); |
| </script><script src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script> |
| <link href="doxygen.css" rel="stylesheet" type="text/css" /> |
| </head> |
| <body> |
| <div id="top"><!-- do not remove this div, it is closed by doxygen! --> |
| <div id="titlearea"> |
| <table cellspacing="0" cellpadding="0"> |
| <tbody> |
| <tr style="height: 56px;"> |
| <td style="padding-left: 0.5em;"> |
| <div id="projectname">Compute Library |
|  <span id="projectnumber">17.10</span> |
| </div> |
| </td> |
| </tr> |
| </tbody> |
| </table> |
| </div> |
| <!-- end header part --> |
| <!-- Generated by Doxygen 1.8.6 --> |
| <script type="text/javascript"> |
| var searchBox = new SearchBox("searchBox", "search",false,'Search'); |
| </script> |
| <div id="navrow1" class="tabs"> |
| <ul class="tablist"> |
| <li><a href="index.xhtml"><span>Main Page</span></a></li> |
| <li><a href="pages.xhtml"><span>Related Pages</span></a></li> |
| <li><a href="namespaces.xhtml"><span>Namespaces</span></a></li> |
| <li><a href="annotated.xhtml"><span>Data Structures</span></a></li> |
| <li class="current"><a href="files.xhtml"><span>Files</span></a></li> |
| <li> |
| <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 id="navrow2" class="tabs2"> |
| <ul class="tablist"> |
| <li><a href="files.xhtml"><span>File List</span></a></li> |
| <li><a href="globals.xhtml"><span>Globals</span></a></li> |
| </ul> |
| </div> |
| </div><!-- top --> |
| <div id="side-nav" class="ui-resizable side-nav-resizable"> |
| <div id="nav-tree"> |
| <div id="nav-tree-contents"> |
| <div id="nav-sync" class="sync"></div> |
| </div> |
| </div> |
| <div id="splitbar" style="-moz-user-select:none;" |
| class="ui-resizable-handle"> |
| </div> |
| </div> |
| <script type="text/javascript"> |
| $(document).ready(function(){initNavTree('gemm_8cl_source.xhtml','');}); |
| </script> |
| <div id="doc-content"> |
| <!-- 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>Namespaces</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(3)"><span class="SelectionMark"> </span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(4)"><span class="SelectionMark"> </span>Functions</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(5)"><span class="SelectionMark"> </span>Variables</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(6)"><span class="SelectionMark"> </span>Typedefs</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(7)"><span class="SelectionMark"> </span>Enumerations</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(8)"><span class="SelectionMark"> </span>Enumerator</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(9)"><span class="SelectionMark"> </span>Friends</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(10)"><span class="SelectionMark"> </span>Macros</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(11)"><span class="SelectionMark"> </span>Pages</a></div> |
| |
| <!-- iframe showing the search results (closed by default) --> |
| <div id="MSearchResultsWindow"> |
| <iframe src="javascript:void(0)" frameborder="0" |
| name="MSearchResults" id="MSearchResults"> |
| </iframe> |
| </div> |
| |
| <div class="header"> |
| <div class="headertitle"> |
| <div class="title">gemm.cl</div> </div> |
| </div><!--header--> |
| <div class="contents"> |
| <a href="gemm_8cl.xhtml">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span> <span class="comment">/*</span></div> |
| <div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment"> * Copyright (c) 2017 ARM Limited.</span></div> |
| <div class="line"><a name="l00003"></a><span class="lineno"> 3</span> <span class="comment"> *</span></div> |
| <div class="line"><a name="l00004"></a><span class="lineno"> 4</span> <span class="comment"> * SPDX-License-Identifier: MIT</span></div> |
| <div class="line"><a name="l00005"></a><span class="lineno"> 5</span> <span class="comment"> *</span></div> |
| <div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div> |
| <div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment"> * of this software and associated documentation files (the "Software"), to</span></div> |
| <div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment"> * deal in the Software without restriction, including without limitation the</span></div> |
| <div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div> |
| <div class="line"><a name="l00010"></a><span class="lineno"> 10</span> <span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div> |
| <div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment"> * furnished to do so, subject to the following conditions:</span></div> |
| <div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment"> *</span></div> |
| <div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div> |
| <div class="line"><a name="l00014"></a><span class="lineno"> 14</span> <span class="comment"> * copies or substantial portions of the Software.</span></div> |
| <div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment"> *</span></div> |
| <div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment"> * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div> |
| <div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div> |
| <div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div> |
| <div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div> |
| <div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div> |
| <div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div> |
| <div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment"> * SOFTWARE.</span></div> |
| <div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment"> */</span></div> |
| <div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="preprocessor">#include "<a class="code" href="helpers_8h.xhtml">helpers.h</a>"</span></div> |
| <div class="line"><a name="l00025"></a><span class="lineno"> 25</span> </div> |
| <div class="line"><a name="l00026"></a><span class="lineno"> 26</span> <span class="preprocessor">#ifdef FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <span class="preprocessor"></span><span class="preprocessor">#include "<a class="code" href="fixed__point_8h.xhtml">fixed_point.h</a>"</span></div> |
| <div class="line"><a name="l00028"></a><span class="lineno"> 28</span> <span class="preprocessor">#endif // FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00029"></a><span class="lineno"> 29</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00045"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#ac5eba21b9ac015e1454044aeba93546c"> 45</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#ac5eba21b9ac015e1454044aeba93546c">gemm_transpose1x4</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00046"></a><span class="lineno"> 46</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00047"></a><span class="lineno"> 47</span> {</div> |
| <div class="line"><a name="l00048"></a><span class="lineno"> 48</span>  uint x = get_global_id(0);</div> |
| <div class="line"><a name="l00049"></a><span class="lineno"> 49</span>  uint y = get_global_id(1);</div> |
| <div class="line"><a name="l00050"></a><span class="lineno"> 50</span> </div> |
| <div class="line"><a name="l00051"></a><span class="lineno"> 51</span>  <span class="comment">// Compute address for Matrix B - source</span></div> |
| <div class="line"><a name="l00052"></a><span class="lineno"> 52</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00053"></a><span class="lineno"> 53</span> </div> |
| <div class="line"><a name="l00054"></a><span class="lineno"> 54</span>  <span class="comment">// Compute address for Matrix B transposed - destination. X and Y are swapped</span></div> |
| <div class="line"><a name="l00055"></a><span class="lineno"> 55</span>  uint dst_addr_in_bytes = y * 16 + ((x * dst_stride_y + dst_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00056"></a><span class="lineno"> 56</span> </div> |
| <div class="line"><a name="l00057"></a><span class="lineno"> 57</span>  uint4 b0 = vload4(0, (__global uint *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l00058"></a><span class="lineno"> 58</span> </div> |
| <div class="line"><a name="l00059"></a><span class="lineno"> 59</span>  vstore4(b0, 0, (__global uint *)(dst_ptr + dst_addr_in_bytes));</div> |
| <div class="line"><a name="l00060"></a><span class="lineno"> 60</span> }</div> |
| <div class="line"><a name="l00061"></a><span class="lineno"> 61</span> </div> |
| <div class="line"><a name="l00077"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#a75c35ac120f123e9cce56c43ba3f48d2"> 77</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#a75c35ac120f123e9cce56c43ba3f48d2">gemm_transpose1x8</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00078"></a><span class="lineno"> 78</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00079"></a><span class="lineno"> 79</span> {</div> |
| <div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  uint x = get_global_id(0);</div> |
| <div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  uint y = get_global_id(1);</div> |
| <div class="line"><a name="l00082"></a><span class="lineno"> 82</span> </div> |
| <div class="line"><a name="l00083"></a><span class="lineno"> 83</span>  <span class="comment">/* Compute address for Matrix B - source */</span></div> |
| <div class="line"><a name="l00084"></a><span class="lineno"> 84</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00085"></a><span class="lineno"> 85</span> </div> |
| <div class="line"><a name="l00086"></a><span class="lineno"> 86</span>  <span class="comment">/* Compute address for Matrix B transposed - destination. X and Y are swapped */</span></div> |
| <div class="line"><a name="l00087"></a><span class="lineno"> 87</span>  uint dst_addr_in_bytes = y * 16 + ((x * dst_stride_y + dst_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00088"></a><span class="lineno"> 88</span> </div> |
| <div class="line"><a name="l00089"></a><span class="lineno"> 89</span>  ushort8 b0 = vload8(0, (__global ushort *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l00090"></a><span class="lineno"> 90</span> </div> |
| <div class="line"><a name="l00091"></a><span class="lineno"> 91</span>  vstore8(b0, 0, (__global ushort *)(dst_ptr + dst_addr_in_bytes));</div> |
| <div class="line"><a name="l00092"></a><span class="lineno"> 92</span> }</div> |
| <div class="line"><a name="l00093"></a><span class="lineno"> 93</span> </div> |
| <div class="line"><a name="l00109"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#a97a299ffb620aa98fe5e39bb8ab0833f"> 109</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#a97a299ffb620aa98fe5e39bb8ab0833f">gemm_transpose1x16</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00110"></a><span class="lineno"> 110</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00111"></a><span class="lineno"> 111</span> {</div> |
| <div class="line"><a name="l00112"></a><span class="lineno"> 112</span>  uint x = get_global_id(0);</div> |
| <div class="line"><a name="l00113"></a><span class="lineno"> 113</span>  uint y = get_global_id(1);</div> |
| <div class="line"><a name="l00114"></a><span class="lineno"> 114</span> </div> |
| <div class="line"><a name="l00115"></a><span class="lineno"> 115</span>  <span class="comment">/* Compute address for Matrix B - source */</span></div> |
| <div class="line"><a name="l00116"></a><span class="lineno"> 116</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00117"></a><span class="lineno"> 117</span> </div> |
| <div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  <span class="comment">/* Compute address for Matrix B transposed - destination. X and Y are swapped */</span></div> |
| <div class="line"><a name="l00119"></a><span class="lineno"> 119</span>  uint dst_addr_in_bytes = y * 16 + ((x * dst_stride_y + dst_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00120"></a><span class="lineno"> 120</span> </div> |
| <div class="line"><a name="l00121"></a><span class="lineno"> 121</span>  uchar16 b0 = vload16(0, (__global uchar *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l00122"></a><span class="lineno"> 122</span> </div> |
| <div class="line"><a name="l00123"></a><span class="lineno"> 123</span>  vstore16(b0, 0, (__global uchar *)(dst_ptr + dst_addr_in_bytes));</div> |
| <div class="line"><a name="l00124"></a><span class="lineno"> 124</span> }</div> |
| <div class="line"><a name="l00125"></a><span class="lineno"> 125</span> </div> |
| <div class="line"><a name="l00141"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#a560b2d50aa886edc0f5daf4fe729717f"> 141</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#a560b2d50aa886edc0f5daf4fe729717f">gemm_interleave4x4_32bit</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00142"></a><span class="lineno"> 142</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00143"></a><span class="lineno"> 143</span> {</div> |
| <div class="line"><a name="l00144"></a><span class="lineno"> 144</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l00145"></a><span class="lineno"> 145</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00146"></a><span class="lineno"> 146</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00147"></a><span class="lineno"> 147</span> </div> |
| <div class="line"><a name="l00148"></a><span class="lineno"> 148</span>  <span class="comment">/* Load values from Matrix A */</span></div> |
| <div class="line"><a name="l00149"></a><span class="lineno"> 149</span>  uint4 a0 = vload4(0, (__global uint *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0)));</div> |
| <div class="line"><a name="l00150"></a><span class="lineno"> 150</span>  uint4 a1 = vload4(0, (__global uint *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 1)));</div> |
| <div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  uint4 a2 = vload4(0, (__global uint *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 2)));</div> |
| <div class="line"><a name="l00152"></a><span class="lineno"> 152</span>  uint4 a3 = vload4(0, (__global uint *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 3)));</div> |
| <div class="line"><a name="l00153"></a><span class="lineno"> 153</span> </div> |
| <div class="line"><a name="l00154"></a><span class="lineno"> 154</span>  uint4 val0 = (uint4)(a0.s0, a1.s0, a2.s0, a3.s0);</div> |
| <div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  vstore4(val0, 0, ((__global uint *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 0);</div> |
| <div class="line"><a name="l00156"></a><span class="lineno"> 156</span> </div> |
| <div class="line"><a name="l00157"></a><span class="lineno"> 157</span>  val0 = (uint4)(a0.s1, a1.s1, a2.s1, a3.s1);</div> |
| <div class="line"><a name="l00158"></a><span class="lineno"> 158</span>  vstore4(val0, 0, ((__global uint *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 4);</div> |
| <div class="line"><a name="l00159"></a><span class="lineno"> 159</span> </div> |
| <div class="line"><a name="l00160"></a><span class="lineno"> 160</span>  val0 = (uint4)(a0.s2, a1.s2, a2.s2, a3.s2);</div> |
| <div class="line"><a name="l00161"></a><span class="lineno"> 161</span>  vstore4(val0, 0, ((__global uint *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 8);</div> |
| <div class="line"><a name="l00162"></a><span class="lineno"> 162</span> </div> |
| <div class="line"><a name="l00163"></a><span class="lineno"> 163</span>  val0 = (uint4)(a0.s3, a1.s3, a2.s3, a3.s3);</div> |
| <div class="line"><a name="l00164"></a><span class="lineno"> 164</span>  vstore4(val0, 0, ((__global uint *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 12);</div> |
| <div class="line"><a name="l00165"></a><span class="lineno"> 165</span> }</div> |
| <div class="line"><a name="l00166"></a><span class="lineno"> 166</span> </div> |
| <div class="line"><a name="l00182"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#a2f32d740e780059f68da8aa589ed0a5b"> 182</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#a2f32d740e780059f68da8aa589ed0a5b">gemm_interleave4x4_16bit</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00183"></a><span class="lineno"> 183</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00184"></a><span class="lineno"> 184</span> {</div> |
| <div class="line"><a name="l00185"></a><span class="lineno"> 185</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l00186"></a><span class="lineno"> 186</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00187"></a><span class="lineno"> 187</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00188"></a><span class="lineno"> 188</span> </div> |
| <div class="line"><a name="l00189"></a><span class="lineno"> 189</span>  <span class="comment">/* Load values from Matrix A */</span></div> |
| <div class="line"><a name="l00190"></a><span class="lineno"> 190</span>  ushort8 a0 = vload8(0, (__global ushort *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0)));</div> |
| <div class="line"><a name="l00191"></a><span class="lineno"> 191</span>  ushort8 a1 = vload8(0, (__global ushort *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 1)));</div> |
| <div class="line"><a name="l00192"></a><span class="lineno"> 192</span>  ushort8 a2 = vload8(0, (__global ushort *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 2)));</div> |
| <div class="line"><a name="l00193"></a><span class="lineno"> 193</span>  ushort8 a3 = vload8(0, (__global ushort *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 3)));</div> |
| <div class="line"><a name="l00194"></a><span class="lineno"> 194</span> </div> |
| <div class="line"><a name="l00195"></a><span class="lineno"> 195</span>  ushort8 val0 = (ushort8)((ushort4)(a0.s0, a1.s0, a2.s0, a3.s0), (ushort4)(a0.s1, a1.s1, a2.s1, a3.s1));</div> |
| <div class="line"><a name="l00196"></a><span class="lineno"> 196</span>  vstore8(val0, 0, ((__global ushort *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 0);</div> |
| <div class="line"><a name="l00197"></a><span class="lineno"> 197</span> </div> |
| <div class="line"><a name="l00198"></a><span class="lineno"> 198</span>  val0 = (ushort8)((ushort4)(a0.s2, a1.s2, a2.s2, a3.s2), (ushort4)(a0.s3, a1.s3, a2.s3, a3.s3));</div> |
| <div class="line"><a name="l00199"></a><span class="lineno"> 199</span>  vstore8(val0, 0, ((__global ushort *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 8);</div> |
| <div class="line"><a name="l00200"></a><span class="lineno"> 200</span> </div> |
| <div class="line"><a name="l00201"></a><span class="lineno"> 201</span>  val0 = (ushort8)((ushort4)(a0.s4, a1.s4, a2.s4, a3.s4), (ushort4)(a0.s5, a1.s5, a2.s5, a3.s5));</div> |
| <div class="line"><a name="l00202"></a><span class="lineno"> 202</span>  vstore8(val0, 0, ((__global ushort *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 16);</div> |
| <div class="line"><a name="l00203"></a><span class="lineno"> 203</span> </div> |
| <div class="line"><a name="l00204"></a><span class="lineno"> 204</span>  val0 = (ushort8)((ushort4)(a0.s6, a1.s6, a2.s6, a3.s6), (ushort4)(a0.s7, a1.s7, a2.s7, a3.s7));</div> |
| <div class="line"><a name="l00205"></a><span class="lineno"> 205</span>  vstore8(val0, 0, ((__global ushort *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 24);</div> |
| <div class="line"><a name="l00206"></a><span class="lineno"> 206</span> }</div> |
| <div class="line"><a name="l00207"></a><span class="lineno"> 207</span> </div> |
| <div class="line"><a name="l00223"></a><span class="lineno"><a class="line" href="gemm_8cl.xhtml#aa66a7b6b0420e54ec173743e6c5bfc45"> 223</a></span> __kernel <span class="keywordtype">void</span> <a class="code" href="gemm_8cl.xhtml#aa66a7b6b0420e54ec173743e6c5bfc45">gemm_interleave4x4_8bit</a>(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l00224"></a><span class="lineno"> 224</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00225"></a><span class="lineno"> 225</span> {</div> |
| <div class="line"><a name="l00226"></a><span class="lineno"> 226</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l00227"></a><span class="lineno"> 227</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l00228"></a><span class="lineno"> 228</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00229"></a><span class="lineno"> 229</span> </div> |
| <div class="line"><a name="l00230"></a><span class="lineno"> 230</span>  <span class="comment">/* Load values from Matrix A */</span></div> |
| <div class="line"><a name="l00231"></a><span class="lineno"> 231</span>  uchar16 a0 = vload16(0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 0)));</div> |
| <div class="line"><a name="l00232"></a><span class="lineno"> 232</span>  uchar16 a1 = vload16(0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 1)));</div> |
| <div class="line"><a name="l00233"></a><span class="lineno"> 233</span>  uchar16 a2 = vload16(0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 2)));</div> |
| <div class="line"><a name="l00234"></a><span class="lineno"> 234</span>  uchar16 a3 = vload16(0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&src, 0, 3)));</div> |
| <div class="line"><a name="l00235"></a><span class="lineno"> 235</span> </div> |
| <div class="line"><a name="l00236"></a><span class="lineno"> 236</span>  uchar16 val0 = (uchar16)((uchar4)(a0.s0, a1.s0, a2.s0, a3.s0), (uchar4)(a0.s1, a1.s1, a2.s1, a3.s1),</div> |
| <div class="line"><a name="l00237"></a><span class="lineno"> 237</span>  (uchar4)(a0.s2, a1.s2, a2.s2, a3.s2), (uchar4)(a0.s3, a1.s3, a2.s3, a3.s3));</div> |
| <div class="line"><a name="l00238"></a><span class="lineno"> 238</span>  vstore16(val0, 0, ((__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 0);</div> |
| <div class="line"><a name="l00239"></a><span class="lineno"> 239</span> </div> |
| <div class="line"><a name="l00240"></a><span class="lineno"> 240</span>  val0 = (uchar16)((uchar4)(a0.s4, a1.s4, a2.s4, a3.s4), (uchar4)(a0.s5, a1.s5, a2.s5, a3.s5),</div> |
| <div class="line"><a name="l00241"></a><span class="lineno"> 241</span>  (uchar4)(a0.s6, a1.s6, a2.s6, a3.s6), (uchar4)(a0.s7, a1.s7, a2.s7, a3.s7));</div> |
| <div class="line"><a name="l00242"></a><span class="lineno"> 242</span>  vstore16(val0, 0, ((__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 16);</div> |
| <div class="line"><a name="l00243"></a><span class="lineno"> 243</span> </div> |
| <div class="line"><a name="l00244"></a><span class="lineno"> 244</span>  val0 = (uchar16)((uchar4)(a0.s8, a1.s8, a2.s8, a3.s8), (uchar4)(a0.s9, a1.s9, a2.s9, a3.s9),</div> |
| <div class="line"><a name="l00245"></a><span class="lineno"> 245</span>  (uchar4)(a0.sA, a1.sA, a2.sA, a3.sA), (uchar4)(a0.sB, a1.sB, a2.sB, a3.sB));</div> |
| <div class="line"><a name="l00246"></a><span class="lineno"> 246</span>  vstore16(val0, 0, ((__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 32);</div> |
| <div class="line"><a name="l00247"></a><span class="lineno"> 247</span> </div> |
| <div class="line"><a name="l00248"></a><span class="lineno"> 248</span>  val0 = (uchar16)((uchar4)(a0.sC, a1.sC, a2.sC, a3.sC), (uchar4)(a0.sD, a1.sD, a2.sD, a3.sD),</div> |
| <div class="line"><a name="l00249"></a><span class="lineno"> 249</span>  (uchar4)(a0.sE, a1.sE, a2.sE, a3.sE), (uchar4)(a0.sF, a1.sF, a2.sF, a3.sF));</div> |
| <div class="line"><a name="l00250"></a><span class="lineno"> 250</span>  vstore16(val0, 0, ((__global uchar *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>) + 48);</div> |
| <div class="line"><a name="l00251"></a><span class="lineno"> 251</span> }</div> |
| <div class="line"><a name="l00252"></a><span class="lineno"> 252</span> </div> |
| <div class="line"><a name="l00268"></a><span class="lineno"> 268</span> <span class="preprocessor">#ifdef DATA_TYPE</span></div> |
| <div class="line"><a name="l00269"></a><span class="lineno"> 269</span> <span class="preprocessor"></span>__kernel <span class="keywordtype">void</span> gemm_accumulate_biases(</div> |
| <div class="line"><a name="l00270"></a><span class="lineno"> 270</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(accum),</div> |
| <div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <a class="code" href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a>(biases))</div> |
| <div class="line"><a name="l00272"></a><span class="lineno"> 272</span> {</div> |
| <div class="line"><a name="l00273"></a><span class="lineno"> 273</span>  <a class="code" href="struct_image.xhtml">Image</a> accum = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(accum);</div> |
| <div class="line"><a name="l00274"></a><span class="lineno"> 274</span>  <a class="code" href="struct_vector.xhtml">Vector</a> biases = <a class="code" href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a>(biases);</div> |
| <div class="line"><a name="l00275"></a><span class="lineno"> 275</span> </div> |
| <div class="line"><a name="l00276"></a><span class="lineno"> 276</span>  <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 16)</div> |
| <div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  accum_value = vload16(0, (__global <a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a> *)accum.ptr);</div> |
| <div class="line"><a name="l00278"></a><span class="lineno"> 278</span>  <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>, 16)</div> |
| <div class="line"><a name="l00279"></a><span class="lineno"> 279</span>  biases_value = vload16(0, (__global DATA_TYPE *)biases.ptr);</div> |
| <div class="line"><a name="l00280"></a><span class="lineno"> 280</span> <span class="preprocessor">#ifdef FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00281"></a><span class="lineno"> 281</span> <span class="preprocessor"></span> accum_value = <a class="code" href="fixed__point_8h.xhtml#a6b1acbaff6cb3bad4edda9a93dac1f9e">ADD_SAT_OP_EXPAND</a>(biases_value, accum_value, DATA_TYPE, 16);</div> |
| <div class="line"><a name="l00282"></a><span class="lineno"> 282</span> <span class="preprocessor">#else // FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00283"></a><span class="lineno"> 283</span> <span class="preprocessor"></span> accum_value = biases_value + accum_value;</div> |
| <div class="line"><a name="l00284"></a><span class="lineno"> 284</span> <span class="preprocessor">#endif // FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00285"></a><span class="lineno"> 285</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00286"></a><span class="lineno"> 286</span>  <span class="comment">// Store result in the accummulate buffer</span></div> |
| <div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  vstore16(accum_value, 0, (__global DATA_TYPE *)accum.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l00288"></a><span class="lineno"> 288</span> }</div> |
| <div class="line"><a name="l00289"></a><span class="lineno"> 289</span> <span class="preprocessor">#endif </span><span class="comment">/* DATA_TYPE */</span><span class="preprocessor"></span></div> |
| <div class="line"><a name="l00290"></a><span class="lineno"> 290</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00291"></a><span class="lineno"> 291</span> <span class="preprocessor">#ifdef COLS_B</span></div> |
| <div class="line"><a name="l00292"></a><span class="lineno"> 292</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00321"></a><span class="lineno"> 321</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_u8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00322"></a><span class="lineno"> 322</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00323"></a><span class="lineno"> 323</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst),</div> |
| <div class="line"><a name="l00324"></a><span class="lineno"> 324</span>  <span class="keywordtype">int</span> a_offset,</div> |
| <div class="line"><a name="l00325"></a><span class="lineno"> 325</span>  <span class="keywordtype">int</span> b_offset,</div> |
| <div class="line"><a name="l00326"></a><span class="lineno"> 326</span>  <span class="keywordtype">int</span> c_offset,</div> |
| <div class="line"><a name="l00327"></a><span class="lineno"> 327</span>  <span class="keywordtype">int</span> c_mult_int,</div> |
| <div class="line"><a name="l00328"></a><span class="lineno"> 328</span>  <span class="keywordtype">int</span> shift)</div> |
| <div class="line"><a name="l00329"></a><span class="lineno"> 329</span> {</div> |
| <div class="line"><a name="l00330"></a><span class="lineno"> 330</span>  <span class="comment">/* src_addr.s0 = address of matrix A */</span></div> |
| <div class="line"><a name="l00331"></a><span class="lineno"> 331</span>  <span class="comment">/* src_addr.s1 = address of matrix B */</span></div> |
| <div class="line"><a name="l00332"></a><span class="lineno"> 332</span> </div> |
| <div class="line"><a name="l00333"></a><span class="lineno"> 333</span>  <span class="comment">/* Compute address for matrix A and B */</span></div> |
| <div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  int2 src_addr = (int2)(get_global_id(1), get_global_id(0)) * (int2)((src0_stride_y),</div> |
| <div class="line"><a name="l00335"></a><span class="lineno"> 335</span>  (src1_stride_y));</div> |
| <div class="line"><a name="l00336"></a><span class="lineno"> 336</span> </div> |
| <div class="line"><a name="l00337"></a><span class="lineno"> 337</span>  <span class="comment">/* Add offset_first_element_in_bytes */</span></div> |
| <div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  src_addr = src_addr + ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00339"></a><span class="lineno"> 339</span> </div> |
| <div class="line"><a name="l00340"></a><span class="lineno"> 340</span>  <span class="comment">/* Compute end row address for matrix B */</span></div> |
| <div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  <span class="keywordtype">int</span> end_row_mtx_b = src_addr.s1 + COLS_B;</div> |
| <div class="line"><a name="l00342"></a><span class="lineno"> 342</span> </div> |
| <div class="line"><a name="l00343"></a><span class="lineno"> 343</span>  <span class="comment">/* Reset accumulators */</span></div> |
| <div class="line"><a name="l00344"></a><span class="lineno"> 344</span>  int16 c00 = 0.0f;</div> |
| <div class="line"><a name="l00345"></a><span class="lineno"> 345</span>  int16 c10 = 0.0f;</div> |
| <div class="line"><a name="l00346"></a><span class="lineno"> 346</span>  int16 c20 = 0.0f;</div> |
| <div class="line"><a name="l00347"></a><span class="lineno"> 347</span>  int16 c30 = 0.0f;</div> |
| <div class="line"><a name="l00348"></a><span class="lineno"> 348</span> </div> |
| <div class="line"><a name="l00349"></a><span class="lineno"> 349</span>  <span class="keywordflow">for</span>(; src_addr.s1 <= (end_row_mtx_b - 8); src_addr += (int2)(8, 32))</div> |
| <div class="line"><a name="l00350"></a><span class="lineno"> 350</span>  {</div> |
| <div class="line"><a name="l00351"></a><span class="lineno"> 351</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  int8 a0 = (int8)a_offset + convert_int8(vload8(0, ((__global uchar *)src0_ptr) + src_addr.s0));</div> |
| <div class="line"><a name="l00353"></a><span class="lineno"> 353</span>  int16 b0 = (int16)b_offset + convert_int16(vload16(0, ((__global uchar *)src1_ptr) + src_addr.s1));</div> |
| <div class="line"><a name="l00354"></a><span class="lineno"> 354</span> </div> |
| <div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  c00 += (int16)a0.s0 * b0;</div> |
| <div class="line"><a name="l00356"></a><span class="lineno"> 356</span>  c10 += (int16)a0.s1 * b0;</div> |
| <div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  c20 += (int16)a0.s2 * b0;</div> |
| <div class="line"><a name="l00358"></a><span class="lineno"> 358</span>  c30 += (int16)a0.s3 * b0;</div> |
| <div class="line"><a name="l00359"></a><span class="lineno"> 359</span> </div> |
| <div class="line"><a name="l00360"></a><span class="lineno"> 360</span>  int16 b1 = (int16)b_offset + convert_int16(vload16(0, ((__global uchar *)src1_ptr) + src_addr.s1 + 16));</div> |
| <div class="line"><a name="l00361"></a><span class="lineno"> 361</span> </div> |
| <div class="line"><a name="l00362"></a><span class="lineno"> 362</span>  c00 += (int16)a0.s4 * b1;</div> |
| <div class="line"><a name="l00363"></a><span class="lineno"> 363</span>  c10 += (int16)a0.s5 * b1;</div> |
| <div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  c20 += (int16)a0.s6 * b1;</div> |
| <div class="line"><a name="l00365"></a><span class="lineno"> 365</span>  c30 += (int16)a0.s7 * b1;</div> |
| <div class="line"><a name="l00366"></a><span class="lineno"> 366</span>  }</div> |
| <div class="line"><a name="l00367"></a><span class="lineno"> 367</span> </div> |
| <div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  <span class="keywordflow">for</span>(; src_addr.s1 < end_row_mtx_b; src_addr += (int2)(4, 16))</div> |
| <div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  {</div> |
| <div class="line"><a name="l00370"></a><span class="lineno"> 370</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  int4 a0 = (int4)a_offset + convert_int4(vload4(0, ((__global uchar *)src0_ptr) + src_addr.s0));</div> |
| <div class="line"><a name="l00372"></a><span class="lineno"> 372</span>  int16 b0 = (int16)b_offset + convert_int16(vload16(0, ((__global uchar *)src1_ptr) + src_addr.s1));</div> |
| <div class="line"><a name="l00373"></a><span class="lineno"> 373</span> </div> |
| <div class="line"><a name="l00374"></a><span class="lineno"> 374</span>  c00 += (int16)a0.s0 * b0;</div> |
| <div class="line"><a name="l00375"></a><span class="lineno"> 375</span>  c10 += (int16)a0.s1 * b0;</div> |
| <div class="line"><a name="l00376"></a><span class="lineno"> 376</span>  c20 += (int16)a0.s2 * b0;</div> |
| <div class="line"><a name="l00377"></a><span class="lineno"> 377</span>  c30 += (int16)a0.s3 * b0;</div> |
| <div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  }</div> |
| <div class="line"><a name="l00379"></a><span class="lineno"> 379</span> </div> |
| <div class="line"><a name="l00380"></a><span class="lineno"> 380</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l00381"></a><span class="lineno"> 381</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00382"></a><span class="lineno"> 382</span> </div> |
| <div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  <span class="comment">/* Multiply by the weight of matrix product */</span></div> |
| <div class="line"><a name="l00384"></a><span class="lineno"> 384</span>  c00 = (((int16)c_offset + c00) * (int16)c_mult_int) >> shift;</div> |
| <div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  c10 = (((int16)c_offset + c10) * (int16)c_mult_int) >> shift;</div> |
| <div class="line"><a name="l00386"></a><span class="lineno"> 386</span>  c20 = (((int16)c_offset + c20) * (int16)c_mult_int) >> shift;</div> |
| <div class="line"><a name="l00387"></a><span class="lineno"> 387</span>  c30 = (((int16)c_offset + c30) * (int16)c_mult_int) >> shift;</div> |
| <div class="line"><a name="l00388"></a><span class="lineno"> 388</span> </div> |
| <div class="line"><a name="l00389"></a><span class="lineno"> 389</span>  <span class="comment">/* Store 4x16 block */</span></div> |
| <div class="line"><a name="l00390"></a><span class="lineno"> 390</span>  vstore16(convert_uchar16_sat(c00), 0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00391"></a><span class="lineno"> 391</span>  vstore16(convert_uchar16_sat(c10), 0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00392"></a><span class="lineno"> 392</span>  vstore16(convert_uchar16_sat(c20), 0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00393"></a><span class="lineno"> 393</span>  vstore16(convert_uchar16_sat(c30), 0, (__global uchar *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00394"></a><span class="lineno"> 394</span> }</div> |
| <div class="line"><a name="l00395"></a><span class="lineno"> 395</span> <span class="preprocessor">#endif </span><span class="comment">/* COLS_B */</span><span class="preprocessor"></span></div> |
| <div class="line"><a name="l00396"></a><span class="lineno"> 396</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00397"></a><span class="lineno"> 397</span> <span class="preprocessor">#if defined(COLS_B) && defined(ALPHA)</span></div> |
| <div class="line"><a name="l00398"></a><span class="lineno"> 398</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00422"></a><span class="lineno"> 422</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f32_midgard(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00423"></a><span class="lineno"> 423</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00424"></a><span class="lineno"> 424</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00425"></a><span class="lineno"> 425</span> {</div> |
| <div class="line"><a name="l00426"></a><span class="lineno"> 426</span>  <span class="comment">/* src_addr.s0 = address of matrix A */</span></div> |
| <div class="line"><a name="l00427"></a><span class="lineno"> 427</span>  <span class="comment">/* src_addr.s1 = address of matrix B */</span></div> |
| <div class="line"><a name="l00428"></a><span class="lineno"> 428</span> </div> |
| <div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  <span class="comment">/* Compute address for matrix A and B */</span></div> |
| <div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  int2 src_addr = (int2)(get_global_id(1), get_global_id(0)) * (int2)((src0_stride_y),</div> |
| <div class="line"><a name="l00431"></a><span class="lineno"> 431</span>  (src1_stride_y));</div> |
| <div class="line"><a name="l00432"></a><span class="lineno"> 432</span> </div> |
| <div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  <span class="comment">/* Add offset_first_element_in_bytes */</span></div> |
| <div class="line"><a name="l00434"></a><span class="lineno"> 434</span>  src_addr = src_addr + ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00435"></a><span class="lineno"> 435</span> </div> |
| <div class="line"><a name="l00436"></a><span class="lineno"> 436</span>  <span class="comment">/* Divide by 4 in order to get the src_addr in unit of float */</span></div> |
| <div class="line"><a name="l00437"></a><span class="lineno"> 437</span>  src_addr = src_addr >> 2;</div> |
| <div class="line"><a name="l00438"></a><span class="lineno"> 438</span> </div> |
| <div class="line"><a name="l00439"></a><span class="lineno"> 439</span>  <span class="comment">/* Compute end row address for matrix B */</span></div> |
| <div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  <span class="keywordtype">int</span> end_row_mtx_b = src_addr.s1 + COLS_B;</div> |
| <div class="line"><a name="l00441"></a><span class="lineno"> 441</span> </div> |
| <div class="line"><a name="l00442"></a><span class="lineno"> 442</span>  <span class="comment">/* Reset accumulators */</span></div> |
| <div class="line"><a name="l00443"></a><span class="lineno"> 443</span>  float4 c00 = 0.0f;</div> |
| <div class="line"><a name="l00444"></a><span class="lineno"> 444</span>  float4 c10 = 0.0f;</div> |
| <div class="line"><a name="l00445"></a><span class="lineno"> 445</span>  float4 c20 = 0.0f;</div> |
| <div class="line"><a name="l00446"></a><span class="lineno"> 446</span>  float4 c30 = 0.0f;</div> |
| <div class="line"><a name="l00447"></a><span class="lineno"> 447</span> </div> |
| <div class="line"><a name="l00448"></a><span class="lineno"> 448</span>  <span class="keywordflow">for</span>(; src_addr.s1 <= (end_row_mtx_b - 8); src_addr += (int2)(8, 8))</div> |
| <div class="line"><a name="l00449"></a><span class="lineno"> 449</span>  {</div> |
| <div class="line"><a name="l00450"></a><span class="lineno"> 450</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00451"></a><span class="lineno"> 451</span>  float4 a0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  float4 b0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00453"></a><span class="lineno"> 453</span> </div> |
| <div class="line"><a name="l00454"></a><span class="lineno"> 454</span>  c00 += (float4)a0.s0 * b0;</div> |
| <div class="line"><a name="l00455"></a><span class="lineno"> 455</span>  c10 += (float4)a0.s1 * b0;</div> |
| <div class="line"><a name="l00456"></a><span class="lineno"> 456</span>  c20 += (float4)a0.s2 * b0;</div> |
| <div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  c30 += (float4)a0.s3 * b0;</div> |
| <div class="line"><a name="l00458"></a><span class="lineno"> 458</span> </div> |
| <div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  a0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src0_ptr) + src_addr.s0 + 4);</div> |
| <div class="line"><a name="l00461"></a><span class="lineno"> 461</span>  b0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src1_ptr) + src_addr.s1 + 4);</div> |
| <div class="line"><a name="l00462"></a><span class="lineno"> 462</span> </div> |
| <div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  c00 += (float4)a0.s0 * b0;</div> |
| <div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  c10 += (float4)a0.s1 * b0;</div> |
| <div class="line"><a name="l00465"></a><span class="lineno"> 465</span>  c20 += (float4)a0.s2 * b0;</div> |
| <div class="line"><a name="l00466"></a><span class="lineno"> 466</span>  c30 += (float4)a0.s3 * b0;</div> |
| <div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  }</div> |
| <div class="line"><a name="l00468"></a><span class="lineno"> 468</span> </div> |
| <div class="line"><a name="l00469"></a><span class="lineno"> 469</span>  <span class="keywordflow">for</span>(; src_addr.s1 < end_row_mtx_b; src_addr += (int2)(4, 4))</div> |
| <div class="line"><a name="l00470"></a><span class="lineno"> 470</span>  {</div> |
| <div class="line"><a name="l00471"></a><span class="lineno"> 471</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00472"></a><span class="lineno"> 472</span>  float4 a0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00473"></a><span class="lineno"> 473</span>  float4 b0 = vload4(0, ((__global <span class="keywordtype">float</span> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00474"></a><span class="lineno"> 474</span> </div> |
| <div class="line"><a name="l00475"></a><span class="lineno"> 475</span>  c00 += (float4)a0.s0 * b0;</div> |
| <div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  c10 += (float4)a0.s1 * b0;</div> |
| <div class="line"><a name="l00477"></a><span class="lineno"> 477</span>  c20 += (float4)a0.s2 * b0;</div> |
| <div class="line"><a name="l00478"></a><span class="lineno"> 478</span>  c30 += (float4)a0.s3 * b0;</div> |
| <div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  }</div> |
| <div class="line"><a name="l00480"></a><span class="lineno"> 480</span> </div> |
| <div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00483"></a><span class="lineno"> 483</span> </div> |
| <div class="line"><a name="l00484"></a><span class="lineno"> 484</span>  <span class="comment">/* Multiply by the weight of matrix product */</span></div> |
| <div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  c00 = c00 * (float4)ALPHA;</div> |
| <div class="line"><a name="l00486"></a><span class="lineno"> 486</span>  c10 = c10 * (float4)ALPHA;</div> |
| <div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  c20 = c20 * (float4)ALPHA;</div> |
| <div class="line"><a name="l00488"></a><span class="lineno"> 488</span>  c30 = c30 * (float4)ALPHA;</div> |
| <div class="line"><a name="l00489"></a><span class="lineno"> 489</span> </div> |
| <div class="line"><a name="l00490"></a><span class="lineno"> 490</span>  <span class="comment">/* Store 4x4 block */</span></div> |
| <div class="line"><a name="l00491"></a><span class="lineno"> 491</span>  vstore4(c00, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00492"></a><span class="lineno"> 492</span>  vstore4(c10, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00493"></a><span class="lineno"> 493</span>  vstore4(c20, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00494"></a><span class="lineno"> 494</span>  vstore4(c30, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00495"></a><span class="lineno"> 495</span> }</div> |
| <div class="line"><a name="l00496"></a><span class="lineno"> 496</span> </div> |
| <div class="line"><a name="l00521"></a><span class="lineno"> 521</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f32_bifrost(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00522"></a><span class="lineno"> 522</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00523"></a><span class="lineno"> 523</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00524"></a><span class="lineno"> 524</span> {</div> |
| <div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  <span class="comment">// src_addr_a = address of matrix A</span></div> |
| <div class="line"><a name="l00526"></a><span class="lineno"> 526</span>  <span class="comment">// src_addr_b = address of matrix B</span></div> |
| <div class="line"><a name="l00527"></a><span class="lineno"> 527</span>  __global <span class="keywordtype">float</span> *src_addr_a = (__global <span class="keywordtype">float</span> *)(src0_ptr + get_global_id(1) * src0_stride_y + src0_offset_first_element_in_bytes);</div> |
| <div class="line"><a name="l00528"></a><span class="lineno"> 528</span>  __global <span class="keywordtype">float</span> *src_addr_b = (__global <span class="keywordtype">float</span> *)(src1_ptr + get_global_id(0) * src1_stride_y + src1_offset_first_element_in_bytes);</div> |
| <div class="line"><a name="l00529"></a><span class="lineno"> 529</span> </div> |
| <div class="line"><a name="l00530"></a><span class="lineno"> 530</span>  <span class="comment">// Compute end row address for matrix B</span></div> |
| <div class="line"><a name="l00531"></a><span class="lineno"> 531</span>  __global <span class="keywordtype">float</span> *src_end_addr_b = src_addr_b + COLS_B;</div> |
| <div class="line"><a name="l00532"></a><span class="lineno"> 532</span> </div> |
| <div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  <span class="comment">// Reset accumulators</span></div> |
| <div class="line"><a name="l00534"></a><span class="lineno"> 534</span>  <span class="keywordtype">float</span> c00 = 0.0f;</div> |
| <div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  <span class="keywordtype">float</span> c01 = 0.0f;</div> |
| <div class="line"><a name="l00536"></a><span class="lineno"> 536</span>  <span class="keywordtype">float</span> c02 = 0.0f;</div> |
| <div class="line"><a name="l00537"></a><span class="lineno"> 537</span>  <span class="keywordtype">float</span> c03 = 0.0f;</div> |
| <div class="line"><a name="l00538"></a><span class="lineno"> 538</span>  <span class="keywordtype">float</span> c10 = 0.0f;</div> |
| <div class="line"><a name="l00539"></a><span class="lineno"> 539</span>  <span class="keywordtype">float</span> c11 = 0.0f;</div> |
| <div class="line"><a name="l00540"></a><span class="lineno"> 540</span>  <span class="keywordtype">float</span> c12 = 0.0f;</div> |
| <div class="line"><a name="l00541"></a><span class="lineno"> 541</span>  <span class="keywordtype">float</span> c13 = 0.0f;</div> |
| <div class="line"><a name="l00542"></a><span class="lineno"> 542</span>  <span class="keywordtype">float</span> c20 = 0.0f;</div> |
| <div class="line"><a name="l00543"></a><span class="lineno"> 543</span>  <span class="keywordtype">float</span> c21 = 0.0f;</div> |
| <div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  <span class="keywordtype">float</span> c22 = 0.0f;</div> |
| <div class="line"><a name="l00545"></a><span class="lineno"> 545</span>  <span class="keywordtype">float</span> c23 = 0.0f;</div> |
| <div class="line"><a name="l00546"></a><span class="lineno"> 546</span>  <span class="keywordtype">float</span> c30 = 0.0f;</div> |
| <div class="line"><a name="l00547"></a><span class="lineno"> 547</span>  <span class="keywordtype">float</span> c31 = 0.0f;</div> |
| <div class="line"><a name="l00548"></a><span class="lineno"> 548</span>  <span class="keywordtype">float</span> c32 = 0.0f;</div> |
| <div class="line"><a name="l00549"></a><span class="lineno"> 549</span>  <span class="keywordtype">float</span> c33 = 0.0f;</div> |
| <div class="line"><a name="l00550"></a><span class="lineno"> 550</span> </div> |
| <div class="line"><a name="l00551"></a><span class="lineno"> 551</span>  <span class="keywordflow">for</span>(; src_addr_b <= (src_end_addr_b - 16); src_addr_a += 16, src_addr_b += 16)</div> |
| <div class="line"><a name="l00552"></a><span class="lineno"> 552</span>  {</div> |
| <div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div> |
| <div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  float4 a0 = vload4(0, src_addr_a);</div> |
| <div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  float4 b0 = vload4(0, src_addr_b);</div> |
| <div class="line"><a name="l00556"></a><span class="lineno"> 556</span> </div> |
| <div class="line"><a name="l00557"></a><span class="lineno"> 557</span>  c00 = fma(a0.s0, b0.s0, c00);</div> |
| <div class="line"><a name="l00558"></a><span class="lineno"> 558</span>  c01 = fma(a0.s0, b0.s1, c01);</div> |
| <div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  c02 = fma(a0.s0, b0.s2, c02);</div> |
| <div class="line"><a name="l00560"></a><span class="lineno"> 560</span>  c03 = fma(a0.s0, b0.s3, c03);</div> |
| <div class="line"><a name="l00561"></a><span class="lineno"> 561</span> </div> |
| <div class="line"><a name="l00562"></a><span class="lineno"> 562</span>  c10 = fma(a0.s1, b0.s0, c10);</div> |
| <div class="line"><a name="l00563"></a><span class="lineno"> 563</span>  c11 = fma(a0.s1, b0.s1, c11);</div> |
| <div class="line"><a name="l00564"></a><span class="lineno"> 564</span>  c12 = fma(a0.s1, b0.s2, c12);</div> |
| <div class="line"><a name="l00565"></a><span class="lineno"> 565</span>  c13 = fma(a0.s1, b0.s3, c13);</div> |
| <div class="line"><a name="l00566"></a><span class="lineno"> 566</span> </div> |
| <div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  c20 = fma(a0.s2, b0.s0, c20);</div> |
| <div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  c21 = fma(a0.s2, b0.s1, c21);</div> |
| <div class="line"><a name="l00569"></a><span class="lineno"> 569</span>  c22 = fma(a0.s2, b0.s2, c22);</div> |
| <div class="line"><a name="l00570"></a><span class="lineno"> 570</span>  c23 = fma(a0.s2, b0.s3, c23);</div> |
| <div class="line"><a name="l00571"></a><span class="lineno"> 571</span> </div> |
| <div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  c30 = fma(a0.s3, b0.s0, c30);</div> |
| <div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  c31 = fma(a0.s3, b0.s1, c31);</div> |
| <div class="line"><a name="l00574"></a><span class="lineno"> 574</span>  c32 = fma(a0.s3, b0.s2, c32);</div> |
| <div class="line"><a name="l00575"></a><span class="lineno"> 575</span>  c33 = fma(a0.s3, b0.s3, c33);</div> |
| <div class="line"><a name="l00576"></a><span class="lineno"> 576</span> </div> |
| <div class="line"><a name="l00577"></a><span class="lineno"> 577</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div> |
| <div class="line"><a name="l00578"></a><span class="lineno"> 578</span>  a0 = vload4(0, src_addr_a + 4);</div> |
| <div class="line"><a name="l00579"></a><span class="lineno"> 579</span>  b0 = vload4(0, src_addr_b + 4);</div> |
| <div class="line"><a name="l00580"></a><span class="lineno"> 580</span> </div> |
| <div class="line"><a name="l00581"></a><span class="lineno"> 581</span>  c00 = fma(a0.s0, b0.s0, c00);</div> |
| <div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  c01 = fma(a0.s0, b0.s1, c01);</div> |
| <div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  c02 = fma(a0.s0, b0.s2, c02);</div> |
| <div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  c03 = fma(a0.s0, b0.s3, c03);</div> |
| <div class="line"><a name="l00585"></a><span class="lineno"> 585</span> </div> |
| <div class="line"><a name="l00586"></a><span class="lineno"> 586</span>  c10 = fma(a0.s1, b0.s0, c10);</div> |
| <div class="line"><a name="l00587"></a><span class="lineno"> 587</span>  c11 = fma(a0.s1, b0.s1, c11);</div> |
| <div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  c12 = fma(a0.s1, b0.s2, c12);</div> |
| <div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  c13 = fma(a0.s1, b0.s3, c13);</div> |
| <div class="line"><a name="l00590"></a><span class="lineno"> 590</span> </div> |
| <div class="line"><a name="l00591"></a><span class="lineno"> 591</span>  c20 = fma(a0.s2, b0.s0, c20);</div> |
| <div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  c21 = fma(a0.s2, b0.s1, c21);</div> |
| <div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  c22 = fma(a0.s2, b0.s2, c22);</div> |
| <div class="line"><a name="l00594"></a><span class="lineno"> 594</span>  c23 = fma(a0.s2, b0.s3, c23);</div> |
| <div class="line"><a name="l00595"></a><span class="lineno"> 595</span> </div> |
| <div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  c30 = fma(a0.s3, b0.s0, c30);</div> |
| <div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  c31 = fma(a0.s3, b0.s1, c31);</div> |
| <div class="line"><a name="l00598"></a><span class="lineno"> 598</span>  c32 = fma(a0.s3, b0.s2, c32);</div> |
| <div class="line"><a name="l00599"></a><span class="lineno"> 599</span>  c33 = fma(a0.s3, b0.s3, c33);</div> |
| <div class="line"><a name="l00600"></a><span class="lineno"> 600</span> </div> |
| <div class="line"><a name="l00601"></a><span class="lineno"> 601</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div> |
| <div class="line"><a name="l00602"></a><span class="lineno"> 602</span>  a0 = vload4(0, src_addr_a + 8);</div> |
| <div class="line"><a name="l00603"></a><span class="lineno"> 603</span>  b0 = vload4(0, src_addr_b + 8);</div> |
| <div class="line"><a name="l00604"></a><span class="lineno"> 604</span> </div> |
| <div class="line"><a name="l00605"></a><span class="lineno"> 605</span>  c00 = fma(a0.s0, b0.s0, c00);</div> |
| <div class="line"><a name="l00606"></a><span class="lineno"> 606</span>  c01 = fma(a0.s0, b0.s1, c01);</div> |
| <div class="line"><a name="l00607"></a><span class="lineno"> 607</span>  c02 = fma(a0.s0, b0.s2, c02);</div> |
| <div class="line"><a name="l00608"></a><span class="lineno"> 608</span>  c03 = fma(a0.s0, b0.s3, c03);</div> |
| <div class="line"><a name="l00609"></a><span class="lineno"> 609</span> </div> |
| <div class="line"><a name="l00610"></a><span class="lineno"> 610</span>  c10 = fma(a0.s1, b0.s0, c10);</div> |
| <div class="line"><a name="l00611"></a><span class="lineno"> 611</span>  c11 = fma(a0.s1, b0.s1, c11);</div> |
| <div class="line"><a name="l00612"></a><span class="lineno"> 612</span>  c12 = fma(a0.s1, b0.s2, c12);</div> |
| <div class="line"><a name="l00613"></a><span class="lineno"> 613</span>  c13 = fma(a0.s1, b0.s3, c13);</div> |
| <div class="line"><a name="l00614"></a><span class="lineno"> 614</span> </div> |
| <div class="line"><a name="l00615"></a><span class="lineno"> 615</span>  c20 = fma(a0.s2, b0.s0, c20);</div> |
| <div class="line"><a name="l00616"></a><span class="lineno"> 616</span>  c21 = fma(a0.s2, b0.s1, c21);</div> |
| <div class="line"><a name="l00617"></a><span class="lineno"> 617</span>  c22 = fma(a0.s2, b0.s2, c22);</div> |
| <div class="line"><a name="l00618"></a><span class="lineno"> 618</span>  c23 = fma(a0.s2, b0.s3, c23);</div> |
| <div class="line"><a name="l00619"></a><span class="lineno"> 619</span> </div> |
| <div class="line"><a name="l00620"></a><span class="lineno"> 620</span>  c30 = fma(a0.s3, b0.s0, c30);</div> |
| <div class="line"><a name="l00621"></a><span class="lineno"> 621</span>  c31 = fma(a0.s3, b0.s1, c31);</div> |
| <div class="line"><a name="l00622"></a><span class="lineno"> 622</span>  c32 = fma(a0.s3, b0.s2, c32);</div> |
| <div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  c33 = fma(a0.s3, b0.s3, c33);</div> |
| <div class="line"><a name="l00624"></a><span class="lineno"> 624</span> </div> |
| <div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div> |
| <div class="line"><a name="l00626"></a><span class="lineno"> 626</span>  a0 = vload4(0, src_addr_a + 12);</div> |
| <div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  b0 = vload4(0, src_addr_b + 12);</div> |
| <div class="line"><a name="l00628"></a><span class="lineno"> 628</span> </div> |
| <div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  c00 = fma(a0.s0, b0.s0, c00);</div> |
| <div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  c01 = fma(a0.s0, b0.s1, c01);</div> |
| <div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  c02 = fma(a0.s0, b0.s2, c02);</div> |
| <div class="line"><a name="l00632"></a><span class="lineno"> 632</span>  c03 = fma(a0.s0, b0.s3, c03);</div> |
| <div class="line"><a name="l00633"></a><span class="lineno"> 633</span> </div> |
| <div class="line"><a name="l00634"></a><span class="lineno"> 634</span>  c10 = fma(a0.s1, b0.s0, c10);</div> |
| <div class="line"><a name="l00635"></a><span class="lineno"> 635</span>  c11 = fma(a0.s1, b0.s1, c11);</div> |
| <div class="line"><a name="l00636"></a><span class="lineno"> 636</span>  c12 = fma(a0.s1, b0.s2, c12);</div> |
| <div class="line"><a name="l00637"></a><span class="lineno"> 637</span>  c13 = fma(a0.s1, b0.s3, c13);</div> |
| <div class="line"><a name="l00638"></a><span class="lineno"> 638</span> </div> |
| <div class="line"><a name="l00639"></a><span class="lineno"> 639</span>  c20 = fma(a0.s2, b0.s0, c20);</div> |
| <div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  c21 = fma(a0.s2, b0.s1, c21);</div> |
| <div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  c22 = fma(a0.s2, b0.s2, c22);</div> |
| <div class="line"><a name="l00642"></a><span class="lineno"> 642</span>  c23 = fma(a0.s2, b0.s3, c23);</div> |
| <div class="line"><a name="l00643"></a><span class="lineno"> 643</span> </div> |
| <div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  c30 = fma(a0.s3, b0.s0, c30);</div> |
| <div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  c31 = fma(a0.s3, b0.s1, c31);</div> |
| <div class="line"><a name="l00646"></a><span class="lineno"> 646</span>  c32 = fma(a0.s3, b0.s2, c32);</div> |
| <div class="line"><a name="l00647"></a><span class="lineno"> 647</span>  c33 = fma(a0.s3, b0.s3, c33);</div> |
| <div class="line"><a name="l00648"></a><span class="lineno"> 648</span>  }</div> |
| <div class="line"><a name="l00649"></a><span class="lineno"> 649</span> </div> |
| <div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  <span class="keywordflow">for</span>(; src_addr_b < src_end_addr_b; src_addr_a += 4, src_addr_b += 4)</div> |
| <div class="line"><a name="l00651"></a><span class="lineno"> 651</span>  {</div> |
| <div class="line"><a name="l00652"></a><span class="lineno"> 652</span>  <span class="comment">// Load values from matrix A (interleaved) and matrix B (transposed)</span></div> |
| <div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  float4 a0 = vload4(0, src_addr_a);</div> |
| <div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  float4 b0 = vload4(0, src_addr_b);</div> |
| <div class="line"><a name="l00655"></a><span class="lineno"> 655</span> </div> |
| <div class="line"><a name="l00656"></a><span class="lineno"> 656</span>  c00 = fma(a0.s0, b0.s0, c00);</div> |
| <div class="line"><a name="l00657"></a><span class="lineno"> 657</span>  c01 = fma(a0.s0, b0.s1, c01);</div> |
| <div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  c02 = fma(a0.s0, b0.s2, c02);</div> |
| <div class="line"><a name="l00659"></a><span class="lineno"> 659</span>  c03 = fma(a0.s0, b0.s3, c03);</div> |
| <div class="line"><a name="l00660"></a><span class="lineno"> 660</span> </div> |
| <div class="line"><a name="l00661"></a><span class="lineno"> 661</span>  c10 = fma(a0.s1, b0.s0, c10);</div> |
| <div class="line"><a name="l00662"></a><span class="lineno"> 662</span>  c11 = fma(a0.s1, b0.s1, c11);</div> |
| <div class="line"><a name="l00663"></a><span class="lineno"> 663</span>  c12 = fma(a0.s1, b0.s2, c12);</div> |
| <div class="line"><a name="l00664"></a><span class="lineno"> 664</span>  c13 = fma(a0.s1, b0.s3, c13);</div> |
| <div class="line"><a name="l00665"></a><span class="lineno"> 665</span> </div> |
| <div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  c20 = fma(a0.s2, b0.s0, c20);</div> |
| <div class="line"><a name="l00667"></a><span class="lineno"> 667</span>  c21 = fma(a0.s2, b0.s1, c21);</div> |
| <div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  c22 = fma(a0.s2, b0.s2, c22);</div> |
| <div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  c23 = fma(a0.s2, b0.s3, c23);</div> |
| <div class="line"><a name="l00670"></a><span class="lineno"> 670</span> </div> |
| <div class="line"><a name="l00671"></a><span class="lineno"> 671</span>  c30 = fma(a0.s3, b0.s0, c30);</div> |
| <div class="line"><a name="l00672"></a><span class="lineno"> 672</span>  c31 = fma(a0.s3, b0.s1, c31);</div> |
| <div class="line"><a name="l00673"></a><span class="lineno"> 673</span>  c32 = fma(a0.s3, b0.s2, c32);</div> |
| <div class="line"><a name="l00674"></a><span class="lineno"> 674</span>  c33 = fma(a0.s3, b0.s3, c33);</div> |
| <div class="line"><a name="l00675"></a><span class="lineno"> 675</span>  }</div> |
| <div class="line"><a name="l00676"></a><span class="lineno"> 676</span> </div> |
| <div class="line"><a name="l00677"></a><span class="lineno"> 677</span>  <span class="comment">// Compute destination address</span></div> |
| <div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00679"></a><span class="lineno"> 679</span> </div> |
| <div class="line"><a name="l00680"></a><span class="lineno"> 680</span>  <span class="comment">// Multiply by the weight of matrix product</span></div> |
| <div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  c00 = c00 * ALPHA;</div> |
| <div class="line"><a name="l00682"></a><span class="lineno"> 682</span>  c01 = c01 * ALPHA;</div> |
| <div class="line"><a name="l00683"></a><span class="lineno"> 683</span>  c02 = c02 * ALPHA;</div> |
| <div class="line"><a name="l00684"></a><span class="lineno"> 684</span>  c03 = c03 * ALPHA;</div> |
| <div class="line"><a name="l00685"></a><span class="lineno"> 685</span>  c10 = c10 * ALPHA;</div> |
| <div class="line"><a name="l00686"></a><span class="lineno"> 686</span>  c11 = c11 * ALPHA;</div> |
| <div class="line"><a name="l00687"></a><span class="lineno"> 687</span>  c12 = c12 * ALPHA;</div> |
| <div class="line"><a name="l00688"></a><span class="lineno"> 688</span>  c13 = c13 * ALPHA;</div> |
| <div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  c20 = c20 * ALPHA;</div> |
| <div class="line"><a name="l00690"></a><span class="lineno"> 690</span>  c21 = c21 * ALPHA;</div> |
| <div class="line"><a name="l00691"></a><span class="lineno"> 691</span>  c22 = c22 * ALPHA;</div> |
| <div class="line"><a name="l00692"></a><span class="lineno"> 692</span>  c23 = c23 * ALPHA;</div> |
| <div class="line"><a name="l00693"></a><span class="lineno"> 693</span>  c30 = c30 * ALPHA;</div> |
| <div class="line"><a name="l00694"></a><span class="lineno"> 694</span>  c31 = c31 * ALPHA;</div> |
| <div class="line"><a name="l00695"></a><span class="lineno"> 695</span>  c32 = c32 * ALPHA;</div> |
| <div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  c33 = c33 * ALPHA;</div> |
| <div class="line"><a name="l00697"></a><span class="lineno"> 697</span> </div> |
| <div class="line"><a name="l00698"></a><span class="lineno"> 698</span>  barrier(CLK_GLOBAL_MEM_FENCE);</div> |
| <div class="line"><a name="l00699"></a><span class="lineno"> 699</span> </div> |
| <div class="line"><a name="l00700"></a><span class="lineno"> 700</span>  <span class="comment">// Store 4x4 block</span></div> |
| <div class="line"><a name="l00701"></a><span class="lineno"> 701</span>  vstore4((float4)(c00, c01, c02, c03), 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00702"></a><span class="lineno"> 702</span>  vstore4((float4)(c10, c11, c12, c13), 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00703"></a><span class="lineno"> 703</span>  vstore4((float4)(c20, c21, c22, c23), 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00704"></a><span class="lineno"> 704</span>  vstore4((float4)(c30, c31, c32, c33), 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00705"></a><span class="lineno"> 705</span> }</div> |
| <div class="line"><a name="l00706"></a><span class="lineno"> 706</span> </div> |
| <div class="line"><a name="l00731"></a><span class="lineno"> 731</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_f16(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00732"></a><span class="lineno"> 732</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00733"></a><span class="lineno"> 733</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00734"></a><span class="lineno"> 734</span> {</div> |
| <div class="line"><a name="l00735"></a><span class="lineno"> 735</span>  <span class="comment">/* src_addr.s0 = address of matrix A */</span></div> |
| <div class="line"><a name="l00736"></a><span class="lineno"> 736</span>  <span class="comment">/* src_addr.s1 = address of matrix B */</span></div> |
| <div class="line"><a name="l00737"></a><span class="lineno"> 737</span> </div> |
| <div class="line"><a name="l00738"></a><span class="lineno"> 738</span>  <span class="comment">/* Compute address for matrix A and B */</span></div> |
| <div class="line"><a name="l00739"></a><span class="lineno"> 739</span>  int2 src_addr = (int2)(get_global_id(1), get_global_id(0)) * (int2)((src0_stride_y),</div> |
| <div class="line"><a name="l00740"></a><span class="lineno"> 740</span>  (src1_stride_y));</div> |
| <div class="line"><a name="l00741"></a><span class="lineno"> 741</span> </div> |
| <div class="line"><a name="l00742"></a><span class="lineno"> 742</span>  <span class="comment">/* Add offset_first_element_in_bytes */</span></div> |
| <div class="line"><a name="l00743"></a><span class="lineno"> 743</span>  src_addr = src_addr + ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00744"></a><span class="lineno"> 744</span> </div> |
| <div class="line"><a name="l00745"></a><span class="lineno"> 745</span>  <span class="comment">/* Divide by 2 in order to get the src_addr in unit of half */</span></div> |
| <div class="line"><a name="l00746"></a><span class="lineno"> 746</span>  src_addr = src_addr >> 1;</div> |
| <div class="line"><a name="l00747"></a><span class="lineno"> 747</span> </div> |
| <div class="line"><a name="l00748"></a><span class="lineno"> 748</span>  <span class="comment">/* Compute end row address for matrix B */</span></div> |
| <div class="line"><a name="l00749"></a><span class="lineno"> 749</span>  <span class="keywordtype">int</span> end_row_mtx_b = src_addr.s1 + COLS_B;</div> |
| <div class="line"><a name="l00750"></a><span class="lineno"> 750</span> </div> |
| <div class="line"><a name="l00751"></a><span class="lineno"> 751</span>  <span class="comment">/* Reset accumulators */</span></div> |
| <div class="line"><a name="l00752"></a><span class="lineno"> 752</span>  half8 c00 = 0.0f;</div> |
| <div class="line"><a name="l00753"></a><span class="lineno"> 753</span>  half8 c10 = 0.0f;</div> |
| <div class="line"><a name="l00754"></a><span class="lineno"> 754</span>  half8 c20 = 0.0f;</div> |
| <div class="line"><a name="l00755"></a><span class="lineno"> 755</span>  half8 c30 = 0.0f;</div> |
| <div class="line"><a name="l00756"></a><span class="lineno"> 756</span> </div> |
| <div class="line"><a name="l00757"></a><span class="lineno"> 757</span>  <span class="keywordflow">for</span>(; src_addr.s1 <= (end_row_mtx_b - 16); src_addr += (int2)(8, 16))</div> |
| <div class="line"><a name="l00758"></a><span class="lineno"> 758</span>  {</div> |
| <div class="line"><a name="l00759"></a><span class="lineno"> 759</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00760"></a><span class="lineno"> 760</span>  half4 a0 = vload4(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00761"></a><span class="lineno"> 761</span>  half8 b0 = vload8(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00762"></a><span class="lineno"> 762</span> </div> |
| <div class="line"><a name="l00763"></a><span class="lineno"> 763</span>  c00 += (half8)a0.s0 * b0;</div> |
| <div class="line"><a name="l00764"></a><span class="lineno"> 764</span>  c10 += (half8)a0.s1 * b0;</div> |
| <div class="line"><a name="l00765"></a><span class="lineno"> 765</span>  c20 += (half8)a0.s2 * b0;</div> |
| <div class="line"><a name="l00766"></a><span class="lineno"> 766</span>  c30 += (half8)a0.s3 * b0;</div> |
| <div class="line"><a name="l00767"></a><span class="lineno"> 767</span> </div> |
| <div class="line"><a name="l00768"></a><span class="lineno"> 768</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00769"></a><span class="lineno"> 769</span>  a0 = vload4(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src0_ptr) + src_addr.s0 + 4);</div> |
| <div class="line"><a name="l00770"></a><span class="lineno"> 770</span>  b0 = vload8(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src1_ptr) + src_addr.s1 + 8);</div> |
| <div class="line"><a name="l00771"></a><span class="lineno"> 771</span> </div> |
| <div class="line"><a name="l00772"></a><span class="lineno"> 772</span>  c00 += (half8)a0.s0 * b0;</div> |
| <div class="line"><a name="l00773"></a><span class="lineno"> 773</span>  c10 += (half8)a0.s1 * b0;</div> |
| <div class="line"><a name="l00774"></a><span class="lineno"> 774</span>  c20 += (half8)a0.s2 * b0;</div> |
| <div class="line"><a name="l00775"></a><span class="lineno"> 775</span>  c30 += (half8)a0.s3 * b0;</div> |
| <div class="line"><a name="l00776"></a><span class="lineno"> 776</span>  }</div> |
| <div class="line"><a name="l00777"></a><span class="lineno"> 777</span> </div> |
| <div class="line"><a name="l00778"></a><span class="lineno"> 778</span>  <span class="keywordflow">for</span>(; src_addr.s1 < end_row_mtx_b; src_addr += (int2)(4, 8))</div> |
| <div class="line"><a name="l00779"></a><span class="lineno"> 779</span>  {</div> |
| <div class="line"><a name="l00780"></a><span class="lineno"> 780</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00781"></a><span class="lineno"> 781</span>  half4 a0 = vload4(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00782"></a><span class="lineno"> 782</span>  half8 b0 = vload8(0, ((__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00783"></a><span class="lineno"> 783</span> </div> |
| <div class="line"><a name="l00784"></a><span class="lineno"> 784</span>  c00 += (half8)a0.s0 * b0;</div> |
| <div class="line"><a name="l00785"></a><span class="lineno"> 785</span>  c10 += (half8)a0.s1 * b0;</div> |
| <div class="line"><a name="l00786"></a><span class="lineno"> 786</span>  c20 += (half8)a0.s2 * b0;</div> |
| <div class="line"><a name="l00787"></a><span class="lineno"> 787</span>  c30 += (half8)a0.s3 * b0;</div> |
| <div class="line"><a name="l00788"></a><span class="lineno"> 788</span>  }</div> |
| <div class="line"><a name="l00789"></a><span class="lineno"> 789</span> </div> |
| <div class="line"><a name="l00790"></a><span class="lineno"> 790</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l00791"></a><span class="lineno"> 791</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00792"></a><span class="lineno"> 792</span> </div> |
| <div class="line"><a name="l00793"></a><span class="lineno"> 793</span>  <span class="comment">/* Multiply by the weight of matrix product */</span></div> |
| <div class="line"><a name="l00794"></a><span class="lineno"> 794</span>  c00 = c00 * (half8)ALPHA;</div> |
| <div class="line"><a name="l00795"></a><span class="lineno"> 795</span>  c10 = c10 * (half8)ALPHA;</div> |
| <div class="line"><a name="l00796"></a><span class="lineno"> 796</span>  c20 = c20 * (half8)ALPHA;</div> |
| <div class="line"><a name="l00797"></a><span class="lineno"> 797</span>  c30 = c30 * (half8)ALPHA;</div> |
| <div class="line"><a name="l00798"></a><span class="lineno"> 798</span> </div> |
| <div class="line"><a name="l00799"></a><span class="lineno"> 799</span>  <span class="comment">/* Store 4x8 block */</span></div> |
| <div class="line"><a name="l00800"></a><span class="lineno"> 800</span>  vstore8(c00, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00801"></a><span class="lineno"> 801</span>  vstore8(c10, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00802"></a><span class="lineno"> 802</span>  vstore8(c20, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00803"></a><span class="lineno"> 803</span>  vstore8(c30, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00804"></a><span class="lineno"> 804</span> }</div> |
| <div class="line"><a name="l00805"></a><span class="lineno"> 805</span> </div> |
| <div class="line"><a name="l00806"></a><span class="lineno"> 806</span> <span class="preprocessor">#ifdef FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l00807"></a><span class="lineno"> 807</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00833"></a><span class="lineno"> 833</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_qs8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00834"></a><span class="lineno"> 834</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00836"></a><span class="lineno"> 836</span> {</div> |
| <div class="line"><a name="l00837"></a><span class="lineno"> 837</span>  <span class="comment">/* src_addr.s0 = address of matrix A */</span></div> |
| <div class="line"><a name="l00838"></a><span class="lineno"> 838</span>  <span class="comment">/* src_addr.s1 = address of matrix B */</span></div> |
| <div class="line"><a name="l00839"></a><span class="lineno"> 839</span> </div> |
| <div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  <span class="comment">/* Compute address for matrix A and B */</span></div> |
| <div class="line"><a name="l00841"></a><span class="lineno"> 841</span>  int2 src_addr = (int2)(get_global_id(1), get_global_id(0)) * (int2)((src0_stride_y),</div> |
| <div class="line"><a name="l00842"></a><span class="lineno"> 842</span>  (src1_stride_y));</div> |
| <div class="line"><a name="l00843"></a><span class="lineno"> 843</span> </div> |
| <div class="line"><a name="l00844"></a><span class="lineno"> 844</span>  <span class="comment">/* Add offset_first_element_in_bytes */</span></div> |
| <div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  src_addr = src_addr + ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00846"></a><span class="lineno"> 846</span> </div> |
| <div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  <span class="comment">/* Compute end row address for matrix B */</span></div> |
| <div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  <span class="keywordtype">int</span> end_row_mtx_b = src_addr.s1 + COLS_B;</div> |
| <div class="line"><a name="l00849"></a><span class="lineno"> 849</span> </div> |
| <div class="line"><a name="l00850"></a><span class="lineno"> 850</span>  <span class="comment">/* Reset accumulators */</span></div> |
| <div class="line"><a name="l00851"></a><span class="lineno"> 851</span>  short8 c00 = 0.0f;</div> |
| <div class="line"><a name="l00852"></a><span class="lineno"> 852</span>  short8 c10 = 0.0f;</div> |
| <div class="line"><a name="l00853"></a><span class="lineno"> 853</span>  short8 c20 = 0.0f;</div> |
| <div class="line"><a name="l00854"></a><span class="lineno"> 854</span>  short8 c30 = 0.0f;</div> |
| <div class="line"><a name="l00855"></a><span class="lineno"> 855</span>  short8 c01 = 0.0f;</div> |
| <div class="line"><a name="l00856"></a><span class="lineno"> 856</span>  short8 c11 = 0.0f;</div> |
| <div class="line"><a name="l00857"></a><span class="lineno"> 857</span>  short8 c21 = 0.0f;</div> |
| <div class="line"><a name="l00858"></a><span class="lineno"> 858</span>  short8 c31 = 0.0f;</div> |
| <div class="line"><a name="l00859"></a><span class="lineno"> 859</span> </div> |
| <div class="line"><a name="l00860"></a><span class="lineno"> 860</span>  <span class="comment">/* This for loop performs 1 accumulation for each iteration */</span></div> |
| <div class="line"><a name="l00861"></a><span class="lineno"> 861</span>  <span class="keywordflow">for</span>(; src_addr.s1 <= (end_row_mtx_b - 16); src_addr += (int2)(4, 16))</div> |
| <div class="line"><a name="l00862"></a><span class="lineno"> 862</span>  {</div> |
| <div class="line"><a name="l00863"></a><span class="lineno"> 863</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00864"></a><span class="lineno"> 864</span>  char4 a0 = vload4(0, ((__global <span class="keywordtype">char</span> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00865"></a><span class="lineno"> 865</span>  char16 b0 = vload16(0, ((__global <span class="keywordtype">char</span> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00866"></a><span class="lineno"> 866</span> </div> |
| <div class="line"><a name="l00867"></a><span class="lineno"> 867</span>  c00 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c00, (char8)a0.s0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00868"></a><span class="lineno"> 868</span>  c10 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c10, (char8)a0.s1, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00869"></a><span class="lineno"> 869</span>  c20 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c20, (char8)a0.s2, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00870"></a><span class="lineno"> 870</span>  c30 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c30, (char8)a0.s3, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00871"></a><span class="lineno"> 871</span> </div> |
| <div class="line"><a name="l00872"></a><span class="lineno"> 872</span>  c01 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c01, (char8)a0.s0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00873"></a><span class="lineno"> 873</span>  c11 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c11, (char8)a0.s1, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00874"></a><span class="lineno"> 874</span>  c21 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c21, (char8)a0.s2, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  c31 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(c31, (char8)a0.s3, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00876"></a><span class="lineno"> 876</span>  }</div> |
| <div class="line"><a name="l00877"></a><span class="lineno"> 877</span> </div> |
| <div class="line"><a name="l00878"></a><span class="lineno"> 878</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l00879"></a><span class="lineno"> 879</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00880"></a><span class="lineno"> 880</span> </div> |
| <div class="line"><a name="l00881"></a><span class="lineno"> 881</span>  <span class="comment">/* Multiply by the weight of matrix product */</span></div> |
| <div class="line"><a name="l00882"></a><span class="lineno"> 882</span>  char16 c00_qs8 = convert_char16_sat((short16)(c00, c01));</div> |
| <div class="line"><a name="l00883"></a><span class="lineno"> 883</span>  char16 c10_qs8 = convert_char16_sat((short16)(c10, c11));</div> |
| <div class="line"><a name="l00884"></a><span class="lineno"> 884</span>  char16 c20_qs8 = convert_char16_sat((short16)(c20, c21));</div> |
| <div class="line"><a name="l00885"></a><span class="lineno"> 885</span>  char16 c30_qs8 = convert_char16_sat((short16)(c30, c31));</div> |
| <div class="line"><a name="l00886"></a><span class="lineno"> 886</span> </div> |
| <div class="line"><a name="l00887"></a><span class="lineno"> 887</span>  c00_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(c00_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00888"></a><span class="lineno"> 888</span>  c10_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(c10_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00889"></a><span class="lineno"> 889</span>  c20_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(c20_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00890"></a><span class="lineno"> 890</span>  c30_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(c30_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00891"></a><span class="lineno"> 891</span> </div> |
| <div class="line"><a name="l00892"></a><span class="lineno"> 892</span>  <span class="comment">/* Store 16x4 block */</span></div> |
| <div class="line"><a name="l00893"></a><span class="lineno"> 893</span>  vstore16(c00_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00894"></a><span class="lineno"> 894</span>  vstore16(c10_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00895"></a><span class="lineno"> 895</span>  vstore16(c20_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00896"></a><span class="lineno"> 896</span>  vstore16(c30_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00897"></a><span class="lineno"> 897</span> }</div> |
| <div class="line"><a name="l00898"></a><span class="lineno"> 898</span> </div> |
| <div class="line"><a name="l00925"></a><span class="lineno"> 925</span> __kernel <span class="keywordtype">void</span> gemm_mm_interleaved_transposed_qs16(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l00927"></a><span class="lineno"> 927</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l00928"></a><span class="lineno"> 928</span> {</div> |
| <div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  <span class="comment">/* src_addr.s0 = address of matrix A */</span></div> |
| <div class="line"><a name="l00930"></a><span class="lineno"> 930</span>  <span class="comment">/* src_addr.s1 = address of matrix B */</span></div> |
| <div class="line"><a name="l00931"></a><span class="lineno"> 931</span> </div> |
| <div class="line"><a name="l00932"></a><span class="lineno"> 932</span>  <span class="comment">/* Compute address for matrix A and B */</span></div> |
| <div class="line"><a name="l00933"></a><span class="lineno"> 933</span>  int2 src_addr = (int2)(get_global_id(1), get_global_id(0)) * (int2)((src0_stride_y),</div> |
| <div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  (src1_stride_y));</div> |
| <div class="line"><a name="l00935"></a><span class="lineno"> 935</span> </div> |
| <div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  <span class="comment">/* Add offset_first_element_in_bytes */</span></div> |
| <div class="line"><a name="l00937"></a><span class="lineno"> 937</span>  src_addr = src_addr + ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l00938"></a><span class="lineno"> 938</span> </div> |
| <div class="line"><a name="l00939"></a><span class="lineno"> 939</span>  <span class="comment">/* Divide by 2 in order to get the src_addr in unit of short */</span></div> |
| <div class="line"><a name="l00940"></a><span class="lineno"> 940</span>  src_addr = src_addr >> 1;</div> |
| <div class="line"><a name="l00941"></a><span class="lineno"> 941</span> </div> |
| <div class="line"><a name="l00942"></a><span class="lineno"> 942</span>  <span class="comment">/* Compute end row address for matrix B */</span></div> |
| <div class="line"><a name="l00943"></a><span class="lineno"> 943</span>  <span class="keywordtype">int</span> end_row_mtx_b = src_addr.s1 + COLS_B;</div> |
| <div class="line"><a name="l00944"></a><span class="lineno"> 944</span> </div> |
| <div class="line"><a name="l00945"></a><span class="lineno"> 945</span>  <span class="comment">/* Reset accumulators */</span></div> |
| <div class="line"><a name="l00946"></a><span class="lineno"> 946</span>  int8 c00 = 0.0f;</div> |
| <div class="line"><a name="l00947"></a><span class="lineno"> 947</span>  int8 c10 = 0.0f;</div> |
| <div class="line"><a name="l00948"></a><span class="lineno"> 948</span>  int8 c20 = 0.0f;</div> |
| <div class="line"><a name="l00949"></a><span class="lineno"> 949</span>  int8 c30 = 0.0f;</div> |
| <div class="line"><a name="l00950"></a><span class="lineno"> 950</span> </div> |
| <div class="line"><a name="l00951"></a><span class="lineno"> 951</span>  <span class="comment">/* This for loop performs 1 accumulation for each iteration */</span></div> |
| <div class="line"><a name="l00952"></a><span class="lineno"> 952</span>  <span class="keywordflow">for</span>(; src_addr.s1 <= (end_row_mtx_b - 8); src_addr += (int2)(4, 8))</div> |
| <div class="line"><a name="l00953"></a><span class="lineno"> 953</span>  {</div> |
| <div class="line"><a name="l00954"></a><span class="lineno"> 954</span>  <span class="comment">/* Load values from matrix A (interleaved) and matrix B (transposed) */</span></div> |
| <div class="line"><a name="l00955"></a><span class="lineno"> 955</span>  short4 a0 = vload4(0, ((__global <span class="keywordtype">short</span> *)src0_ptr) + src_addr.s0);</div> |
| <div class="line"><a name="l00956"></a><span class="lineno"> 956</span>  short8 b0 = vload8(0, ((__global <span class="keywordtype">short</span> *)src1_ptr) + src_addr.s1);</div> |
| <div class="line"><a name="l00957"></a><span class="lineno"> 957</span> </div> |
| <div class="line"><a name="l00958"></a><span class="lineno"> 958</span>  c00 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(c00, (short8)a0.s0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00959"></a><span class="lineno"> 959</span>  c10 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(c10, (short8)a0.s1, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00960"></a><span class="lineno"> 960</span>  c20 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(c20, (short8)a0.s2, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00961"></a><span class="lineno"> 961</span>  c30 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(c30, (short8)a0.s3, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00962"></a><span class="lineno"> 962</span>  }</div> |
| <div class="line"><a name="l00963"></a><span class="lineno"> 963</span> </div> |
| <div class="line"><a name="l00964"></a><span class="lineno"> 964</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l00965"></a><span class="lineno"> 965</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l00966"></a><span class="lineno"> 966</span> </div> |
| <div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  <span class="comment">/* Multiply by the weight of matrix product */</span></div> |
| <div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  short8 c00_qs16 = convert_short8_sat(c00);</div> |
| <div class="line"><a name="l00969"></a><span class="lineno"> 969</span>  short8 c10_qs16 = convert_short8_sat(c10);</div> |
| <div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  short8 c20_qs16 = convert_short8_sat(c20);</div> |
| <div class="line"><a name="l00971"></a><span class="lineno"> 971</span>  short8 c30_qs16 = convert_short8_sat(c30);</div> |
| <div class="line"><a name="l00972"></a><span class="lineno"> 972</span> </div> |
| <div class="line"><a name="l00973"></a><span class="lineno"> 973</span>  c00_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(c00_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00974"></a><span class="lineno"> 974</span>  c10_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(c10_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00975"></a><span class="lineno"> 975</span>  c20_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(c20_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00976"></a><span class="lineno"> 976</span>  c30_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(c30_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l00977"></a><span class="lineno"> 977</span> </div> |
| <div class="line"><a name="l00978"></a><span class="lineno"> 978</span>  <span class="comment">/* Store 8x4 block */</span></div> |
| <div class="line"><a name="l00979"></a><span class="lineno"> 979</span>  vstore8(c00_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l00980"></a><span class="lineno"> 980</span>  vstore8(c10_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l00981"></a><span class="lineno"> 981</span>  vstore8(c20_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l00982"></a><span class="lineno"> 982</span>  vstore8(c30_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l00983"></a><span class="lineno"> 983</span> }</div> |
| <div class="line"><a name="l00984"></a><span class="lineno"> 984</span> <span class="preprocessor">#endif // defined(FIXED_POINT_POSITION)</span></div> |
| <div class="line"><a name="l00985"></a><span class="lineno"> 985</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l00986"></a><span class="lineno"> 986</span> <span class="preprocessor">#if defined(COLS_A) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && (NUM_ELEMS_PROCESSED_PER_THREAD_Y)</span></div> |
| <div class="line"><a name="l00987"></a><span class="lineno"> 987</span> <span class="preprocessor"></span><span class="preprocessor">#if defined(DATA_TYPE)</span></div> |
| <div class="line"><a name="l00988"></a><span class="lineno"> 988</span> <span class="preprocessor"></span><span class="preprocessor">#define VECTOR_TYPE VEC_DATA_TYPE(DATA_TYPE, NUM_ELEMS_PROCESSED_PER_THREAD_X)</span></div> |
| <div class="line"><a name="l00989"></a><span class="lineno"> 989</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01015"></a><span class="lineno"> 1015</span> __kernel <span class="keywordtype">void</span> gemm_mm_floating_point(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01018"></a><span class="lineno"> 1018</span> {</div> |
| <div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div> |
| <div class="line"><a name="l01020"></a><span class="lineno"> 1020</span> </div> |
| <div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div> |
| <div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l01023"></a><span class="lineno"> 1023</span> </div> |
| <div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>  <span class="comment">// Update address for the matrix A</span></div> |
| <div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div> |
| <div class="line"><a name="l01026"></a><span class="lineno"> 1026</span> </div> |
| <div class="line"><a name="l01027"></a><span class="lineno"> 1027</span>  <span class="comment">// Update address for the matrix B</span></div> |
| <div class="line"><a name="l01028"></a><span class="lineno"> 1028</span>  src_addr.s1 += idx * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>);</div> |
| <div class="line"><a name="l01029"></a><span class="lineno"> 1029</span> </div> |
| <div class="line"><a name="l01030"></a><span class="lineno"> 1030</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (COLS_A * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>));</div> |
| <div class="line"><a name="l01031"></a><span class="lineno"> 1031</span> </div> |
| <div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>  VECTOR_TYPE acc0 = 0.0f;</div> |
| <div class="line"><a name="l01033"></a><span class="lineno"> 1033</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01034"></a><span class="lineno"> 1034</span> <span class="preprocessor"></span> VECTOR_TYPE acc1 = 0.0f;</div> |
| <div class="line"><a name="l01035"></a><span class="lineno"> 1035</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01036"></a><span class="lineno"> 1036</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01037"></a><span class="lineno"> 1037</span> <span class="preprocessor"></span> VECTOR_TYPE acc2 = 0.0f;</div> |
| <div class="line"><a name="l01038"></a><span class="lineno"> 1038</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01039"></a><span class="lineno"> 1039</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01040"></a><span class="lineno"> 1040</span> <span class="preprocessor"></span> VECTOR_TYPE acc3 = 0.0f;</div> |
| <div class="line"><a name="l01041"></a><span class="lineno"> 1041</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01042"></a><span class="lineno"> 1042</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01043"></a><span class="lineno"> 1043</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 2 * <span class="keyword">sizeof</span>(<a class="code" href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a>)); src_addr += (int2)(2 * <span class="keyword">sizeof</span>(DATA_TYPE), 2 * src1_stride_y))</div> |
| <div class="line"><a name="l01044"></a><span class="lineno"> 1044</span>  {</div> |
| <div class="line"><a name="l01045"></a><span class="lineno"> 1045</span>  <span class="comment">// Load values from matrix A</span></div> |
| <div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>  <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE, 2)</div> |
| <div class="line"><a name="l01047"></a><span class="lineno"> 1047</span>  a0 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01048"></a><span class="lineno"> 1048</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01049"></a><span class="lineno"> 1049</span> <span class="preprocessor"></span> <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE, 2)</div> |
| <div class="line"><a name="l01050"></a><span class="lineno"> 1050</span>  a1 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01051"></a><span class="lineno"> 1051</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01052"></a><span class="lineno"> 1052</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01053"></a><span class="lineno"> 1053</span> <span class="preprocessor"></span> <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE, 2)</div> |
| <div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>  a2 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01055"></a><span class="lineno"> 1055</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01056"></a><span class="lineno"> 1056</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01057"></a><span class="lineno"> 1057</span> <span class="preprocessor"></span> <a class="code" href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a>(DATA_TYPE, 2)</div> |
| <div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>  a3 = vload2(0, (__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01059"></a><span class="lineno"> 1059</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01060"></a><span class="lineno"> 1060</span> <span class="preprocessor"></span> <span class="comment">// Load values from matrix B</span></div> |
| <div class="line"><a name="l01061"></a><span class="lineno"> 1061</span>  VECTOR_TYPE b0 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01062"></a><span class="lineno"> 1062</span>  VECTOR_TYPE b1 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1 + src1_stride_y));</div> |
| <div class="line"><a name="l01063"></a><span class="lineno"> 1063</span> </div> |
| <div class="line"><a name="l01064"></a><span class="lineno"> 1064</span>  <span class="comment">// Accumulate</span></div> |
| <div class="line"><a name="l01065"></a><span class="lineno"> 1065</span>  acc0 += b0 * (VECTOR_TYPE)a0.s0;</div> |
| <div class="line"><a name="l01066"></a><span class="lineno"> 1066</span>  acc0 += b1 * (VECTOR_TYPE)a0.s1;</div> |
| <div class="line"><a name="l01067"></a><span class="lineno"> 1067</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01068"></a><span class="lineno"> 1068</span> <span class="preprocessor"></span> acc1 += b0 * (VECTOR_TYPE)a1.s0;</div> |
| <div class="line"><a name="l01069"></a><span class="lineno"> 1069</span>  acc1 += b1 * (VECTOR_TYPE)a1.s1;</div> |
| <div class="line"><a name="l01070"></a><span class="lineno"> 1070</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01071"></a><span class="lineno"> 1071</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01072"></a><span class="lineno"> 1072</span> <span class="preprocessor"></span> acc2 += b0 * (VECTOR_TYPE)a2.s0;</div> |
| <div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>  acc2 += b1 * (VECTOR_TYPE)a2.s1;</div> |
| <div class="line"><a name="l01074"></a><span class="lineno"> 1074</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01075"></a><span class="lineno"> 1075</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01076"></a><span class="lineno"> 1076</span> <span class="preprocessor"></span> acc3 += b0 * (VECTOR_TYPE)a3.s0;</div> |
| <div class="line"><a name="l01077"></a><span class="lineno"> 1077</span>  acc3 += b1 * (VECTOR_TYPE)a3.s1;</div> |
| <div class="line"><a name="l01078"></a><span class="lineno"> 1078</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01079"></a><span class="lineno"> 1079</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01080"></a><span class="lineno"> 1080</span> </div> |
| <div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(<span class="keyword">sizeof</span>(DATA_TYPE), src1_stride_y))</div> |
| <div class="line"><a name="l01082"></a><span class="lineno"> 1082</span>  {</div> |
| <div class="line"><a name="l01083"></a><span class="lineno"> 1083</span>  <span class="comment">// Load values from matrix A</span></div> |
| <div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>  DATA_TYPE a0 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01085"></a><span class="lineno"> 1085</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01086"></a><span class="lineno"> 1086</span> <span class="preprocessor"></span> DATA_TYPE a1 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01087"></a><span class="lineno"> 1087</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01088"></a><span class="lineno"> 1088</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01089"></a><span class="lineno"> 1089</span> <span class="preprocessor"></span> DATA_TYPE a2 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01090"></a><span class="lineno"> 1090</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01091"></a><span class="lineno"> 1091</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01092"></a><span class="lineno"> 1092</span> <span class="preprocessor"></span> DATA_TYPE a3 = *((__global DATA_TYPE *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01093"></a><span class="lineno"> 1093</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01094"></a><span class="lineno"> 1094</span> <span class="preprocessor"></span> <span class="comment">// Load values from matrix B</span></div> |
| <div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>  VECTOR_TYPE b0 = <a class="code" href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)(0, (__global DATA_TYPE *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01096"></a><span class="lineno"> 1096</span> </div> |
| <div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>  <span class="comment">// Accumulate</span></div> |
| <div class="line"><a name="l01098"></a><span class="lineno"> 1098</span>  acc0 += b0 * (VECTOR_TYPE)a0;</div> |
| <div class="line"><a name="l01099"></a><span class="lineno"> 1099</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01100"></a><span class="lineno"> 1100</span> <span class="preprocessor"></span> acc1 += b0 * (VECTOR_TYPE)a1;</div> |
| <div class="line"><a name="l01101"></a><span class="lineno"> 1101</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01102"></a><span class="lineno"> 1102</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01103"></a><span class="lineno"> 1103</span> <span class="preprocessor"></span> acc2 += b0 * (VECTOR_TYPE)a2;</div> |
| <div class="line"><a name="l01104"></a><span class="lineno"> 1104</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01105"></a><span class="lineno"> 1105</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01106"></a><span class="lineno"> 1106</span> <span class="preprocessor"></span> acc3 += b0 * (VECTOR_TYPE)a3;</div> |
| <div class="line"><a name="l01107"></a><span class="lineno"> 1107</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01108"></a><span class="lineno"> 1108</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01109"></a><span class="lineno"> 1109</span> </div> |
| <div class="line"><a name="l01110"></a><span class="lineno"> 1110</span>  <span class="comment">// Compute destination address</span></div> |
| <div class="line"><a name="l01111"></a><span class="lineno"> 1111</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01112"></a><span class="lineno"> 1112</span> </div> |
| <div class="line"><a name="l01113"></a><span class="lineno"> 1113</span>  <span class="comment">// Multiply by the weight of matrix-matrix product and store the result</span></div> |
| <div class="line"><a name="l01114"></a><span class="lineno"> 1114</span>  acc0 = acc0 * (VECTOR_TYPE)ALPHA;</div> |
| <div class="line"><a name="l01115"></a><span class="lineno"> 1115</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div> |
| <div class="line"><a name="l01116"></a><span class="lineno"> 1116</span>  (acc0, 0, (__global DATA_TYPE *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l01117"></a><span class="lineno"> 1117</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01118"></a><span class="lineno"> 1118</span> <span class="preprocessor"></span> acc1 = acc1 * (VECTOR_TYPE)ALPHA;</div> |
| <div class="line"><a name="l01119"></a><span class="lineno"> 1119</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div> |
| <div class="line"><a name="l01120"></a><span class="lineno"> 1120</span>  (acc1, 0, (__global DATA_TYPE *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l01121"></a><span class="lineno"> 1121</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01122"></a><span class="lineno"> 1122</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01123"></a><span class="lineno"> 1123</span> <span class="preprocessor"></span> acc2 = acc2 * (VECTOR_TYPE)ALPHA;</div> |
| <div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div> |
| <div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>  (acc2, 0, (__global DATA_TYPE *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l01126"></a><span class="lineno"> 1126</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01127"></a><span class="lineno"> 1127</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01128"></a><span class="lineno"> 1128</span> <span class="preprocessor"></span> acc3 = acc3 * (VECTOR_TYPE)ALPHA;</div> |
| <div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>  <a class="code" href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a>(NUM_ELEMS_PROCESSED_PER_THREAD_X)</div> |
| <div class="line"><a name="l01130"></a><span class="lineno"> 1130</span>  (acc3, 0, (__global DATA_TYPE *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l01131"></a><span class="lineno"> 1131</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01132"></a><span class="lineno"> 1132</span> <span class="preprocessor"></span>}</div> |
| <div class="line"><a name="l01133"></a><span class="lineno"> 1133</span> <span class="preprocessor">#endif // defined(DATA_TYPE)</span></div> |
| <div class="line"><a name="l01134"></a><span class="lineno"> 1134</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01135"></a><span class="lineno"> 1135</span> <span class="preprocessor">#ifdef FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l01136"></a><span class="lineno"> 1136</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01163"></a><span class="lineno"> 1163</span> __kernel <span class="keywordtype">void</span> gemm_mm_qs8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01166"></a><span class="lineno"> 1166</span> {</div> |
| <div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div> |
| <div class="line"><a name="l01168"></a><span class="lineno"> 1168</span> </div> |
| <div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div> |
| <div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l01171"></a><span class="lineno"> 1171</span> </div> |
| <div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>  <span class="comment">// Update address for the matrix A</span></div> |
| <div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div> |
| <div class="line"><a name="l01174"></a><span class="lineno"> 1174</span> </div> |
| <div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>  <span class="comment">// Update address for the matrix B</span></div> |
| <div class="line"><a name="l01176"></a><span class="lineno"> 1176</span>  src_addr.s1 += idx * <span class="keyword">sizeof</span>(char);</div> |
| <div class="line"><a name="l01177"></a><span class="lineno"> 1177</span> </div> |
| <div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (COLS_A * <span class="keyword">sizeof</span>(char));</div> |
| <div class="line"><a name="l01179"></a><span class="lineno"> 1179</span> </div> |
| <div class="line"><a name="l01180"></a><span class="lineno"> 1180</span>  short8 acc00 = 0;</div> |
| <div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>  short8 acc01 = 0;</div> |
| <div class="line"><a name="l01182"></a><span class="lineno"> 1182</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01183"></a><span class="lineno"> 1183</span> <span class="preprocessor"></span> short8 acc10 = 0;</div> |
| <div class="line"><a name="l01184"></a><span class="lineno"> 1184</span>  short8 acc11 = 0;</div> |
| <div class="line"><a name="l01185"></a><span class="lineno"> 1185</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01186"></a><span class="lineno"> 1186</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01187"></a><span class="lineno"> 1187</span> <span class="preprocessor"></span> short8 acc20 = 0;</div> |
| <div class="line"><a name="l01188"></a><span class="lineno"> 1188</span>  short8 acc21 = 0;</div> |
| <div class="line"><a name="l01189"></a><span class="lineno"> 1189</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01190"></a><span class="lineno"> 1190</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01191"></a><span class="lineno"> 1191</span> <span class="preprocessor"></span> short8 acc30 = 0;</div> |
| <div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>  short8 acc31 = 0;</div> |
| <div class="line"><a name="l01193"></a><span class="lineno"> 1193</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01194"></a><span class="lineno"> 1194</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01195"></a><span class="lineno"> 1195</span>  <span class="comment">// This for loop performs 4 accumulations per iteration</span></div> |
| <div class="line"><a name="l01196"></a><span class="lineno"> 1196</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 2); src_addr += (int2)(2, 2 * src1_stride_y))</div> |
| <div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>  {</div> |
| <div class="line"><a name="l01198"></a><span class="lineno"> 1198</span>  char2 a0 = vload2(0, (__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01199"></a><span class="lineno"> 1199</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01200"></a><span class="lineno"> 1200</span> <span class="preprocessor"></span> char2 a1 = vload2(0, (__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01201"></a><span class="lineno"> 1201</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01202"></a><span class="lineno"> 1202</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01203"></a><span class="lineno"> 1203</span> <span class="preprocessor"></span> char2 a2 = vload2(0, (__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01204"></a><span class="lineno"> 1204</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01205"></a><span class="lineno"> 1205</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01206"></a><span class="lineno"> 1206</span> <span class="preprocessor"></span> char2 a3 = vload2(0, (__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01207"></a><span class="lineno"> 1207</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01208"></a><span class="lineno"> 1208</span> <span class="preprocessor"></span> char16 b0 = vload16(0, (__global <span class="keywordtype">char</span> *)(src1_ptr + src_addr.s1 + 0 * src1_stride_y));</div> |
| <div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>  char16 b1 = vload16(0, (__global <span class="keywordtype">char</span> *)(src1_ptr + src_addr.s1 + 1 * src1_stride_y));</div> |
| <div class="line"><a name="l01210"></a><span class="lineno"> 1210</span> </div> |
| <div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>  acc00 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc00, (char8)a0.s0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>  acc00 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc00, (char8)a0.s1, b1.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01213"></a><span class="lineno"> 1213</span>  acc01 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc01, (char8)a0.s0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01214"></a><span class="lineno"> 1214</span>  acc01 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc01, (char8)a0.s1, b1.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01215"></a><span class="lineno"> 1215</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01216"></a><span class="lineno"> 1216</span> <span class="preprocessor"></span> acc10 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc10, (char8)a1.s0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>  acc10 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc10, (char8)a1.s1, b1.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>  acc11 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc11, (char8)a1.s0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>  acc11 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc11, (char8)a1.s1, b1.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01220"></a><span class="lineno"> 1220</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01221"></a><span class="lineno"> 1221</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01222"></a><span class="lineno"> 1222</span> <span class="preprocessor"></span> acc20 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc20, (char8)a2.s0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01223"></a><span class="lineno"> 1223</span>  acc20 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc20, (char8)a2.s1, b1.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>  acc21 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc21, (char8)a2.s0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>  acc21 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc21, (char8)a2.s1, b1.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01226"></a><span class="lineno"> 1226</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01227"></a><span class="lineno"> 1227</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01228"></a><span class="lineno"> 1228</span> <span class="preprocessor"></span> acc30 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc30, (char8)a3.s0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>  acc30 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc30, (char8)a3.s1, b1.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>  acc31 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc31, (char8)a3.s0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01231"></a><span class="lineno"> 1231</span>  acc31 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc31, (char8)a3.s1, b1.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01232"></a><span class="lineno"> 1232</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01233"></a><span class="lineno"> 1233</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01234"></a><span class="lineno"> 1234</span> </div> |
| <div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>  <span class="comment">// Left-over accumulations</span></div> |
| <div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(1, src1_stride_y))</div> |
| <div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>  {</div> |
| <div class="line"><a name="l01238"></a><span class="lineno"> 1238</span>  <span class="keywordtype">char</span> a0 = *((__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01239"></a><span class="lineno"> 1239</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01240"></a><span class="lineno"> 1240</span> <span class="preprocessor"></span> <span class="keywordtype">char</span> a1 = *((__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01241"></a><span class="lineno"> 1241</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01242"></a><span class="lineno"> 1242</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01243"></a><span class="lineno"> 1243</span> <span class="preprocessor"></span> <span class="keywordtype">char</span> a2 = *((__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01244"></a><span class="lineno"> 1244</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01245"></a><span class="lineno"> 1245</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01246"></a><span class="lineno"> 1246</span> <span class="preprocessor"></span> <span class="keywordtype">char</span> a3 = *((__global <span class="keywordtype">char</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01247"></a><span class="lineno"> 1247</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01248"></a><span class="lineno"> 1248</span> <span class="preprocessor"></span> char16 b0 = vload16(0, (__global <span class="keywordtype">char</span> *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01249"></a><span class="lineno"> 1249</span> </div> |
| <div class="line"><a name="l01250"></a><span class="lineno"> 1250</span>  acc00 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc00, (char8)a0, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>  acc01 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc01, (char8)a0, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01252"></a><span class="lineno"> 1252</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01253"></a><span class="lineno"> 1253</span> <span class="preprocessor"></span> acc10 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc10, (char8)a1, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01254"></a><span class="lineno"> 1254</span>  acc11 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc11, (char8)a1, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01255"></a><span class="lineno"> 1255</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01256"></a><span class="lineno"> 1256</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01257"></a><span class="lineno"> 1257</span> <span class="preprocessor"></span> acc20 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc20, (char8)a2, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01258"></a><span class="lineno"> 1258</span>  acc21 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc21, (char8)a2, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01259"></a><span class="lineno"> 1259</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01260"></a><span class="lineno"> 1260</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01261"></a><span class="lineno"> 1261</span> <span class="preprocessor"></span> acc30 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc30, (char8)a3, b0.s01234567, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>  acc31 = <a class="code" href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a>(acc31, (char8)a3, b0.s89ABCDEF, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01263"></a><span class="lineno"> 1263</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01264"></a><span class="lineno"> 1264</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01265"></a><span class="lineno"> 1265</span> </div> |
| <div class="line"><a name="l01266"></a><span class="lineno"> 1266</span>  <span class="comment">// Compute destination address</span></div> |
| <div class="line"><a name="l01267"></a><span class="lineno"> 1267</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01268"></a><span class="lineno"> 1268</span> </div> |
| <div class="line"><a name="l01269"></a><span class="lineno"> 1269</span>  <span class="comment">// Multiply by the weight of matrix product and store the result</span></div> |
| <div class="line"><a name="l01270"></a><span class="lineno"> 1270</span>  char16 acc_qs8;</div> |
| <div class="line"><a name="l01271"></a><span class="lineno"> 1271</span>  acc_qs8 = convert_char16_sat((short16)(acc00, acc01));</div> |
| <div class="line"><a name="l01272"></a><span class="lineno"> 1272</span>  acc_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(acc_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01273"></a><span class="lineno"> 1273</span>  vstore16(acc_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l01274"></a><span class="lineno"> 1274</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01275"></a><span class="lineno"> 1275</span> <span class="preprocessor"></span> acc_qs8 = convert_char16_sat((short16)(acc10, acc11));</div> |
| <div class="line"><a name="l01276"></a><span class="lineno"> 1276</span>  acc_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(acc_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01277"></a><span class="lineno"> 1277</span>  vstore16(acc_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l01278"></a><span class="lineno"> 1278</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01279"></a><span class="lineno"> 1279</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01280"></a><span class="lineno"> 1280</span> <span class="preprocessor"></span> acc_qs8 = convert_char16_sat((short16)(acc20, acc21));</div> |
| <div class="line"><a name="l01281"></a><span class="lineno"> 1281</span>  acc_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(acc_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01282"></a><span class="lineno"> 1282</span>  vstore16(acc_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l01283"></a><span class="lineno"> 1283</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01284"></a><span class="lineno"> 1284</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01285"></a><span class="lineno"> 1285</span> <span class="preprocessor"></span> acc_qs8 = convert_char16_sat((short16)(acc30, acc31));</div> |
| <div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>  acc_qs8 = <a class="code" href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a>(acc_qs8, (char16)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01287"></a><span class="lineno"> 1287</span>  vstore16(acc_qs8, 0, (__global <span class="keywordtype">char</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l01288"></a><span class="lineno"> 1288</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01289"></a><span class="lineno"> 1289</span> <span class="preprocessor"></span>}</div> |
| <div class="line"><a name="l01290"></a><span class="lineno"> 1290</span> </div> |
| <div class="line"><a name="l01318"></a><span class="lineno"> 1318</span> __kernel <span class="keywordtype">void</span> gemm_mm_qs16(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01321"></a><span class="lineno"> 1321</span> {</div> |
| <div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * NUM_ELEMS_PROCESSED_PER_THREAD_X;</div> |
| <div class="line"><a name="l01323"></a><span class="lineno"> 1323</span> </div> |
| <div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>  <span class="comment">// Compute starting address for matrix A and Matrix B</span></div> |
| <div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes, src1_offset_first_element_in_bytes));</div> |
| <div class="line"><a name="l01326"></a><span class="lineno"> 1326</span> </div> |
| <div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>  <span class="comment">// Update address for the matrix A</span></div> |
| <div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>  src_addr.s0 += get_global_id(1) * src0_stride_y * NUM_ELEMS_PROCESSED_PER_THREAD_Y;</div> |
| <div class="line"><a name="l01329"></a><span class="lineno"> 1329</span> </div> |
| <div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>  <span class="comment">// Update address for the matrix B</span></div> |
| <div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>  src_addr.s1 += idx * <span class="keyword">sizeof</span>(short);</div> |
| <div class="line"><a name="l01332"></a><span class="lineno"> 1332</span> </div> |
| <div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (COLS_A * <span class="keyword">sizeof</span>(short));</div> |
| <div class="line"><a name="l01334"></a><span class="lineno"> 1334</span> </div> |
| <div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>  int8 acc0 = 0;</div> |
| <div class="line"><a name="l01336"></a><span class="lineno"> 1336</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01337"></a><span class="lineno"> 1337</span> <span class="preprocessor"></span> int8 acc1 = 0;</div> |
| <div class="line"><a name="l01338"></a><span class="lineno"> 1338</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01339"></a><span class="lineno"> 1339</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01340"></a><span class="lineno"> 1340</span> <span class="preprocessor"></span> int8 acc2 = 0;</div> |
| <div class="line"><a name="l01341"></a><span class="lineno"> 1341</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01342"></a><span class="lineno"> 1342</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01343"></a><span class="lineno"> 1343</span> <span class="preprocessor"></span> int8 acc3 = 0;</div> |
| <div class="line"><a name="l01344"></a><span class="lineno"> 1344</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01345"></a><span class="lineno"> 1345</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01346"></a><span class="lineno"> 1346</span>  <span class="comment">// This for loop performs 4 accumulations per iteration</span></div> |
| <div class="line"><a name="l01347"></a><span class="lineno"> 1347</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 2 * <span class="keyword">sizeof</span>(short)); src_addr += (int2)(2 * <span class="keyword">sizeof</span>(<span class="keywordtype">short</span>), 2 * src1_stride_y))</div> |
| <div class="line"><a name="l01348"></a><span class="lineno"> 1348</span>  {</div> |
| <div class="line"><a name="l01349"></a><span class="lineno"> 1349</span>  short2 a0 = vload2(0, (__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01350"></a><span class="lineno"> 1350</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01351"></a><span class="lineno"> 1351</span> <span class="preprocessor"></span> short2 a1 = vload2(0, (__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01352"></a><span class="lineno"> 1352</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01353"></a><span class="lineno"> 1353</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01354"></a><span class="lineno"> 1354</span> <span class="preprocessor"></span> short2 a2 = vload2(0, (__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01355"></a><span class="lineno"> 1355</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01356"></a><span class="lineno"> 1356</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01357"></a><span class="lineno"> 1357</span> <span class="preprocessor"></span> short2 a3 = vload2(0, (__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01358"></a><span class="lineno"> 1358</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01359"></a><span class="lineno"> 1359</span> <span class="preprocessor"></span> short8 b0 = vload8(0, (__global <span class="keywordtype">short</span> *)(src1_ptr + src_addr.s1 + 0 * src1_stride_y));</div> |
| <div class="line"><a name="l01360"></a><span class="lineno"> 1360</span>  short8 b1 = vload8(0, (__global <span class="keywordtype">short</span> *)(src1_ptr + src_addr.s1 + 1 * src1_stride_y));</div> |
| <div class="line"><a name="l01361"></a><span class="lineno"> 1361</span> </div> |
| <div class="line"><a name="l01362"></a><span class="lineno"> 1362</span>  acc0 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc0, (short8)a0.s0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01363"></a><span class="lineno"> 1363</span>  acc0 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc0, (short8)a0.s1, b1, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01364"></a><span class="lineno"> 1364</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01365"></a><span class="lineno"> 1365</span> <span class="preprocessor"></span> acc1 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc1, (short8)a1.s0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01366"></a><span class="lineno"> 1366</span>  acc1 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc1, (short8)a1.s1, b1, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01367"></a><span class="lineno"> 1367</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01368"></a><span class="lineno"> 1368</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01369"></a><span class="lineno"> 1369</span> <span class="preprocessor"></span> acc2 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc2, (short8)a2.s0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01370"></a><span class="lineno"> 1370</span>  acc2 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc2, (short8)a2.s1, b1, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01371"></a><span class="lineno"> 1371</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01372"></a><span class="lineno"> 1372</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01373"></a><span class="lineno"> 1373</span> <span class="preprocessor"></span> acc3 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc3, (short8)a3.s0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01374"></a><span class="lineno"> 1374</span>  acc3 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc3, (short8)a3.s1, b1, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01375"></a><span class="lineno"> 1375</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01376"></a><span class="lineno"> 1376</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01377"></a><span class="lineno"> 1377</span> </div> |
| <div class="line"><a name="l01378"></a><span class="lineno"> 1378</span>  <span class="comment">// Left-over accumulations</span></div> |
| <div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(<span class="keyword">sizeof</span>(<span class="keywordtype">short</span>), src1_stride_y))</div> |
| <div class="line"><a name="l01380"></a><span class="lineno"> 1380</span>  {</div> |
| <div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>  <span class="keywordtype">short</span> a0 = *((__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y));</div> |
| <div class="line"><a name="l01382"></a><span class="lineno"> 1382</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01383"></a><span class="lineno"> 1383</span> <span class="preprocessor"></span> <span class="keywordtype">short</span> a1 = *((__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 1 * src0_stride_y));</div> |
| <div class="line"><a name="l01384"></a><span class="lineno"> 1384</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01385"></a><span class="lineno"> 1385</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01386"></a><span class="lineno"> 1386</span> <span class="preprocessor"></span> <span class="keywordtype">short</span> a2 = *((__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 2 * src0_stride_y));</div> |
| <div class="line"><a name="l01387"></a><span class="lineno"> 1387</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01388"></a><span class="lineno"> 1388</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01389"></a><span class="lineno"> 1389</span> <span class="preprocessor"></span> <span class="keywordtype">short</span> a3 = *((__global <span class="keywordtype">short</span> *)(src0_ptr + src_addr.s0 + 3 * src0_stride_y));</div> |
| <div class="line"><a name="l01390"></a><span class="lineno"> 1390</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01391"></a><span class="lineno"> 1391</span> <span class="preprocessor"></span> short8 b0 = vload8(0, (__global <span class="keywordtype">short</span> *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01392"></a><span class="lineno"> 1392</span> </div> |
| <div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>  acc0 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc0, (short8)a0, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01394"></a><span class="lineno"> 1394</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01395"></a><span class="lineno"> 1395</span> <span class="preprocessor"></span> acc1 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc1, (short8)a1, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01396"></a><span class="lineno"> 1396</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01397"></a><span class="lineno"> 1397</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01398"></a><span class="lineno"> 1398</span> <span class="preprocessor"></span> acc2 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc2, (short8)a2, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01399"></a><span class="lineno"> 1399</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01400"></a><span class="lineno"> 1400</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01401"></a><span class="lineno"> 1401</span> <span class="preprocessor"></span> acc3 = <a class="code" href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a>(acc3, (short8)a3, b0, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01402"></a><span class="lineno"> 1402</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01403"></a><span class="lineno"> 1403</span> <span class="preprocessor"></span> }</div> |
| <div class="line"><a name="l01404"></a><span class="lineno"> 1404</span> </div> |
| <div class="line"><a name="l01405"></a><span class="lineno"> 1405</span>  <span class="comment">// Compute destination address</span></div> |
| <div class="line"><a name="l01406"></a><span class="lineno"> 1406</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01407"></a><span class="lineno"> 1407</span> </div> |
| <div class="line"><a name="l01408"></a><span class="lineno"> 1408</span>  <span class="comment">// Multiply by the weight of matrix product and store the result</span></div> |
| <div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>  short8 acc_qs16;</div> |
| <div class="line"><a name="l01410"></a><span class="lineno"> 1410</span>  acc_qs16 = convert_short8_sat(acc0);</div> |
| <div class="line"><a name="l01411"></a><span class="lineno"> 1411</span>  acc_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(acc_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01412"></a><span class="lineno"> 1412</span>  vstore8(acc_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l01413"></a><span class="lineno"> 1413</span> <span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01414"></a><span class="lineno"> 1414</span> <span class="preprocessor"></span> acc_qs16 = convert_short8_sat(acc1);</div> |
| <div class="line"><a name="l01415"></a><span class="lineno"> 1415</span>  acc_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(acc_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01416"></a><span class="lineno"> 1416</span>  vstore8(acc_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 1)));</div> |
| <div class="line"><a name="l01417"></a><span class="lineno"> 1417</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1</span></div> |
| <div class="line"><a name="l01418"></a><span class="lineno"> 1418</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01419"></a><span class="lineno"> 1419</span> <span class="preprocessor"></span> acc_qs16 = convert_short8_sat(acc2);</div> |
| <div class="line"><a name="l01420"></a><span class="lineno"> 1420</span>  acc_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(acc_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01421"></a><span class="lineno"> 1421</span>  vstore8(acc_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 2)));</div> |
| <div class="line"><a name="l01422"></a><span class="lineno"> 1422</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 2</span></div> |
| <div class="line"><a name="l01423"></a><span class="lineno"> 1423</span> <span class="preprocessor"></span><span class="preprocessor">#if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01424"></a><span class="lineno"> 1424</span> <span class="preprocessor"></span> acc_qs16 = convert_short8_sat(acc3);</div> |
| <div class="line"><a name="l01425"></a><span class="lineno"> 1425</span>  acc_qs16 = <a class="code" href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a>(acc_qs16, (short8)ALPHA, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01426"></a><span class="lineno"> 1426</span>  vstore8(acc_qs16, 0, (__global <span class="keywordtype">short</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 3)));</div> |
| <div class="line"><a name="l01427"></a><span class="lineno"> 1427</span> <span class="preprocessor">#endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3</span></div> |
| <div class="line"><a name="l01428"></a><span class="lineno"> 1428</span> <span class="preprocessor"></span>}</div> |
| <div class="line"><a name="l01429"></a><span class="lineno"> 1429</span> <span class="preprocessor">#endif // defined(FIXED_POINT_POSITION)</span></div> |
| <div class="line"><a name="l01430"></a><span class="lineno"> 1430</span> <span class="preprocessor"></span><span class="preprocessor">#endif // defined(COLS_A) && defined(NUM_ELEMS_PROCESSED_PER_THREAD_X) && (NUM_ELEMS_PROCESSED_PER_THREAD_Y)</span></div> |
| <div class="line"><a name="l01431"></a><span class="lineno"> 1431</span> <span class="preprocessor"></span><span class="preprocessor">#endif // defined(COLS_B) && defined(ALPHA)</span></div> |
| <div class="line"><a name="l01432"></a><span class="lineno"> 1432</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01433"></a><span class="lineno"> 1433</span> <span class="preprocessor">#ifdef BETA</span></div> |
| <div class="line"><a name="l01434"></a><span class="lineno"> 1434</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01451"></a><span class="lineno"> 1451</span> __kernel <span class="keywordtype">void</span> gemm_ma_f32(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l01452"></a><span class="lineno"> 1452</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01453"></a><span class="lineno"> 1453</span> {</div> |
| <div class="line"><a name="l01454"></a><span class="lineno"> 1454</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l01455"></a><span class="lineno"> 1455</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l01456"></a><span class="lineno"> 1456</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01457"></a><span class="lineno"> 1457</span> </div> |
| <div class="line"><a name="l01458"></a><span class="lineno"> 1458</span>  <span class="comment">/* Load values from A x B */</span></div> |
| <div class="line"><a name="l01459"></a><span class="lineno"> 1459</span>  float4 alpha_ab = vload4(0, (__global <span class="keywordtype">float</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01460"></a><span class="lineno"> 1460</span> </div> |
| <div class="line"><a name="l01461"></a><span class="lineno"> 1461</span>  <span class="comment">/* Load values from Matrix C */</span></div> |
| <div class="line"><a name="l01462"></a><span class="lineno"> 1462</span>  float4 c = vload4(0, (__global <span class="keywordtype">float</span> *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01463"></a><span class="lineno"> 1463</span> </div> |
| <div class="line"><a name="l01464"></a><span class="lineno"> 1464</span>  <span class="comment">/* Computes alpha * axb + beta * c */</span></div> |
| <div class="line"><a name="l01465"></a><span class="lineno"> 1465</span>  float4 out = alpha_ab + (float4)BETA * c;</div> |
| <div class="line"><a name="l01466"></a><span class="lineno"> 1466</span> </div> |
| <div class="line"><a name="l01467"></a><span class="lineno"> 1467</span>  <span class="comment">/* Store final result in axb matrix */</span></div> |
| <div class="line"><a name="l01468"></a><span class="lineno"> 1468</span>  vstore4(out, 0, (__global <span class="keywordtype">float</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01469"></a><span class="lineno"> 1469</span> }</div> |
| <div class="line"><a name="l01470"></a><span class="lineno"> 1470</span> </div> |
| <div class="line"><a name="l01488"></a><span class="lineno"> 1488</span> __kernel <span class="keywordtype">void</span> gemm_ma_f16(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l01489"></a><span class="lineno"> 1489</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01490"></a><span class="lineno"> 1490</span> {</div> |
| <div class="line"><a name="l01491"></a><span class="lineno"> 1491</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l01492"></a><span class="lineno"> 1492</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l01493"></a><span class="lineno"> 1493</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01494"></a><span class="lineno"> 1494</span> </div> |
| <div class="line"><a name="l01495"></a><span class="lineno"> 1495</span>  <span class="comment">/* Load values from A x B */</span></div> |
| <div class="line"><a name="l01496"></a><span class="lineno"> 1496</span>  half8 alpha_ab = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01497"></a><span class="lineno"> 1497</span> </div> |
| <div class="line"><a name="l01498"></a><span class="lineno"> 1498</span>  <span class="comment">/* Load values from Matrix C */</span></div> |
| <div class="line"><a name="l01499"></a><span class="lineno"> 1499</span>  half8 c = vload8(0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01500"></a><span class="lineno"> 1500</span> </div> |
| <div class="line"><a name="l01501"></a><span class="lineno"> 1501</span>  <span class="comment">/* Computes alpha * axb + beta * c */</span></div> |
| <div class="line"><a name="l01502"></a><span class="lineno"> 1502</span>  half8 out = alpha_ab + (half8)BETA * c;</div> |
| <div class="line"><a name="l01503"></a><span class="lineno"> 1503</span> </div> |
| <div class="line"><a name="l01504"></a><span class="lineno"> 1504</span>  <span class="comment">/* Store final result in axb matrix */</span></div> |
| <div class="line"><a name="l01505"></a><span class="lineno"> 1505</span>  vstore8(out, 0, (__global <a class="code" href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">half</a> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01506"></a><span class="lineno"> 1506</span> }</div> |
| <div class="line"><a name="l01507"></a><span class="lineno"> 1507</span> </div> |
| <div class="line"><a name="l01508"></a><span class="lineno"> 1508</span> <span class="preprocessor">#ifdef FIXED_POINT_POSITION</span></div> |
| <div class="line"><a name="l01509"></a><span class="lineno"> 1509</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01528"></a><span class="lineno"> 1528</span> __kernel <span class="keywordtype">void</span> gemm_ma_qs8(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l01529"></a><span class="lineno"> 1529</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01530"></a><span class="lineno"> 1530</span> {</div> |
| <div class="line"><a name="l01531"></a><span class="lineno"> 1531</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l01532"></a><span class="lineno"> 1532</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l01533"></a><span class="lineno"> 1533</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01534"></a><span class="lineno"> 1534</span> </div> |
| <div class="line"><a name="l01535"></a><span class="lineno"> 1535</span>  <span class="comment">/* Load values from A x B */</span></div> |
| <div class="line"><a name="l01536"></a><span class="lineno"> 1536</span>  char16 alpha_ab = vload16(0, (__global <span class="keywordtype">char</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01537"></a><span class="lineno"> 1537</span> </div> |
| <div class="line"><a name="l01538"></a><span class="lineno"> 1538</span>  <span class="comment">/* Load values from Matrix C */</span></div> |
| <div class="line"><a name="l01539"></a><span class="lineno"> 1539</span>  char16 c = vload16(0, (__global <span class="keywordtype">char</span> *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01540"></a><span class="lineno"> 1540</span> </div> |
| <div class="line"><a name="l01541"></a><span class="lineno"> 1541</span>  <span class="comment">/* Computes alpha * axb + beta * c */</span></div> |
| <div class="line"><a name="l01542"></a><span class="lineno"> 1542</span>  char16 out = <a class="code" href="fixed__point_8h.xhtml#ad59f32bf4f61044a698ff4da02080e5a">mla_sat_qs8x16</a>(alpha_ab, (char16)BETA, c, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01543"></a><span class="lineno"> 1543</span> </div> |
| <div class="line"><a name="l01544"></a><span class="lineno"> 1544</span>  <span class="comment">/* Store final result in axb matrix */</span></div> |
| <div class="line"><a name="l01545"></a><span class="lineno"> 1545</span>  vstore16(out, 0, (__global <span class="keywordtype">char</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01546"></a><span class="lineno"> 1546</span> }</div> |
| <div class="line"><a name="l01547"></a><span class="lineno"> 1547</span> </div> |
| <div class="line"><a name="l01567"></a><span class="lineno"> 1567</span> __kernel <span class="keywordtype">void</span> gemm_ma_qs16(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src),</div> |
| <div class="line"><a name="l01568"></a><span class="lineno"> 1568</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01569"></a><span class="lineno"> 1569</span> {</div> |
| <div class="line"><a name="l01570"></a><span class="lineno"> 1570</span>  <span class="comment">/* Compute source and destination addresses */</span></div> |
| <div class="line"><a name="l01571"></a><span class="lineno"> 1571</span>  <a class="code" href="struct_image.xhtml">Image</a> src = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(src);</div> |
| <div class="line"><a name="l01572"></a><span class="lineno"> 1572</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01573"></a><span class="lineno"> 1573</span> </div> |
| <div class="line"><a name="l01574"></a><span class="lineno"> 1574</span>  <span class="comment">/* Load values from A x B */</span></div> |
| <div class="line"><a name="l01575"></a><span class="lineno"> 1575</span>  short8 alpha_ab = vload8(0, (__global <span class="keywordtype">short</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01576"></a><span class="lineno"> 1576</span> </div> |
| <div class="line"><a name="l01577"></a><span class="lineno"> 1577</span>  <span class="comment">/* Load values from Matrix C */</span></div> |
| <div class="line"><a name="l01578"></a><span class="lineno"> 1578</span>  short8 c = vload8(0, (__global <span class="keywordtype">short</span> *)src.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01579"></a><span class="lineno"> 1579</span> </div> |
| <div class="line"><a name="l01580"></a><span class="lineno"> 1580</span>  <span class="comment">/* Computes alpha * axb + beta * c */</span></div> |
| <div class="line"><a name="l01581"></a><span class="lineno"> 1581</span>  short8 out = <a class="code" href="fixed__point_8h.xhtml#a7a3c295a894e0529251bd4529b36a574">mla_sat_qs16x8</a>(alpha_ab, (short8)BETA, c, FIXED_POINT_POSITION);</div> |
| <div class="line"><a name="l01582"></a><span class="lineno"> 1582</span> </div> |
| <div class="line"><a name="l01583"></a><span class="lineno"> 1583</span>  <span class="comment">/* Store final result in axb matrix */</span></div> |
| <div class="line"><a name="l01584"></a><span class="lineno"> 1584</span>  vstore8(out, 0, (__global <span class="keywordtype">short</span> *)dst.<a class="code" href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">ptr</a>);</div> |
| <div class="line"><a name="l01585"></a><span class="lineno"> 1585</span> }</div> |
| <div class="line"><a name="l01586"></a><span class="lineno"> 1586</span> <span class="preprocessor">#endif </span><span class="comment">/* defined(FIXED_POINT_POSITION) */</span><span class="preprocessor"></span></div> |
| <div class="line"><a name="l01587"></a><span class="lineno"> 1587</span> <span class="preprocessor"></span><span class="preprocessor">#endif </span><span class="comment">/* defined(BETA) */</span><span class="preprocessor"></span></div> |
| <div class="line"><a name="l01588"></a><span class="lineno"> 1588</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01589"></a><span class="lineno"> 1589</span> <span class="preprocessor">#ifdef WIDTH_VECTOR_A</span></div> |
| <div class="line"><a name="l01590"></a><span class="lineno"> 1590</span> <span class="preprocessor"></span></div> |
| <div class="line"><a name="l01617"></a><span class="lineno"> 1617</span> __kernel <span class="keywordtype">void</span> gemm_lc_vm_f32(<a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(src0),</div> |
| <div class="line"><a name="l01618"></a><span class="lineno"> 1618</span>  <a class="code" href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a>(src1),</div> |
| <div class="line"><a name="l01619"></a><span class="lineno"> 1619</span>  <a class="code" href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a>(dst))</div> |
| <div class="line"><a name="l01620"></a><span class="lineno"> 1620</span> {</div> |
| <div class="line"><a name="l01621"></a><span class="lineno"> 1621</span>  <span class="keywordtype">int</span> idx = get_global_id(0) * 4;</div> |
| <div class="line"><a name="l01622"></a><span class="lineno"> 1622</span>  <span class="keywordtype">int</span> idy = get_global_id(1);</div> |
| <div class="line"><a name="l01623"></a><span class="lineno"> 1623</span> </div> |
| <div class="line"><a name="l01624"></a><span class="lineno"> 1624</span>  <span class="comment">/* Compute the address for the vector A and matrix B */</span></div> |
| <div class="line"><a name="l01625"></a><span class="lineno"> 1625</span>  int2 src_addr = ((int2)(src0_offset_first_element_in_bytes + src0_stride_y * idy, src1_offset_first_element_in_bytes + src1_stride_z * idy));</div> |
| <div class="line"><a name="l01626"></a><span class="lineno"> 1626</span>  src_addr.s1 += idx * <span class="keyword">sizeof</span>(float);</div> |
| <div class="line"><a name="l01627"></a><span class="lineno"> 1627</span> </div> |
| <div class="line"><a name="l01628"></a><span class="lineno"> 1628</span>  <span class="keywordtype">int</span> end_row_vec_a = src_addr.s0 + (WIDTH_VECTOR_A * <span class="keyword">sizeof</span>(float));</div> |
| <div class="line"><a name="l01629"></a><span class="lineno"> 1629</span> </div> |
| <div class="line"><a name="l01630"></a><span class="lineno"> 1630</span>  float4 acc = 0.0f;</div> |
| <div class="line"><a name="l01631"></a><span class="lineno"> 1631</span> </div> |
| <div class="line"><a name="l01632"></a><span class="lineno"> 1632</span>  <span class="keywordflow">for</span>(; src_addr.s0 <= (end_row_vec_a - 2 * <span class="keyword">sizeof</span>(float)); src_addr += (int2)(2 * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>), 2 * src1_stride_y))</div> |
| <div class="line"><a name="l01633"></a><span class="lineno"> 1633</span>  {</div> |
| <div class="line"><a name="l01634"></a><span class="lineno"> 1634</span>  float2 a0 = vload2(0, (__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0));</div> |
| <div class="line"><a name="l01635"></a><span class="lineno"> 1635</span>  float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01636"></a><span class="lineno"> 1636</span>  float4 b1 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1 + src1_stride_y));</div> |
| <div class="line"><a name="l01637"></a><span class="lineno"> 1637</span> </div> |
| <div class="line"><a name="l01638"></a><span class="lineno"> 1638</span>  acc += b0 * (float4)a0.s0;</div> |
| <div class="line"><a name="l01639"></a><span class="lineno"> 1639</span>  acc += b1 * (float4)a0.s1;</div> |
| <div class="line"><a name="l01640"></a><span class="lineno"> 1640</span>  }</div> |
| <div class="line"><a name="l01641"></a><span class="lineno"> 1641</span> </div> |
| <div class="line"><a name="l01642"></a><span class="lineno"> 1642</span>  <span class="keywordflow">for</span>(; src_addr.s0 < end_row_vec_a; src_addr += (int2)(<span class="keyword">sizeof</span>(<span class="keywordtype">float</span>), src1_stride_y))</div> |
| <div class="line"><a name="l01643"></a><span class="lineno"> 1643</span>  {</div> |
| <div class="line"><a name="l01644"></a><span class="lineno"> 1644</span>  <span class="keywordtype">float</span> a0 = *((__global <span class="keywordtype">float</span> *)(src0_ptr + src_addr.s0));</div> |
| <div class="line"><a name="l01645"></a><span class="lineno"> 1645</span>  float4 b0 = vload4(0, (__global <span class="keywordtype">float</span> *)(src1_ptr + src_addr.s1));</div> |
| <div class="line"><a name="l01646"></a><span class="lineno"> 1646</span> </div> |
| <div class="line"><a name="l01647"></a><span class="lineno"> 1647</span>  acc += b0 * (float4)a0;</div> |
| <div class="line"><a name="l01648"></a><span class="lineno"> 1648</span>  }</div> |
| <div class="line"><a name="l01649"></a><span class="lineno"> 1649</span> </div> |
| <div class="line"><a name="l01650"></a><span class="lineno"> 1650</span>  <span class="comment">/* Compute destination address */</span></div> |
| <div class="line"><a name="l01651"></a><span class="lineno"> 1651</span>  <a class="code" href="struct_image.xhtml">Image</a> dst = <a class="code" href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a>(dst);</div> |
| <div class="line"><a name="l01652"></a><span class="lineno"> 1652</span> </div> |
| <div class="line"><a name="l01653"></a><span class="lineno"> 1653</span>  vstore4(acc, 0, (__global <span class="keywordtype">float</span> *)(<a class="code" href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a>(&dst, 0, 0)));</div> |
| <div class="line"><a name="l01654"></a><span class="lineno"> 1654</span> }</div> |
| <div class="line"><a name="l01655"></a><span class="lineno"> 1655</span> <span class="preprocessor">#endif </span><span class="comment">/* WIDTH_VECTOR_A */</span><span class="preprocessor"></span></div> |
| <div class="ttc" id="struct_vector_xhtml"><div class="ttname"><a href="struct_vector.xhtml">Vector</a></div><div class="ttdoc">Structure to hold Vector information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00126">helpers.h:126</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a286a2ecb07602fd7268d31c6e88e62a6"><div class="ttname"><a href="fixed__point_8h.xhtml#a286a2ecb07602fd7268d31c6e88e62a6">mul_sat_qs8x16</a></div><div class="ttdeci">qs8x16 mul_sat_qs8x16(qs8x16 VopA, qs8x16 VopB, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00249">fixed_point.h:249</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_a97a299ffb620aa98fe5e39bb8ab0833f"><div class="ttname"><a href="gemm_8cl.xhtml#a97a299ffb620aa98fe5e39bb8ab0833f">gemm_transpose1x16</a></div><div class="ttdeci">__kernel void gemm_transpose1x16(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel computes the "vector" 1x16 transposition of input matrix. </div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00109">gemm.cl:109</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a93fefd4972b6b9440c8371509b46d19e"><div class="ttname"><a href="fixed__point_8h.xhtml#a93fefd4972b6b9440c8371509b46d19e">mul_sat_qs16x8</a></div><div class="ttdeci">qs16x8 mul_sat_qs16x8(qs16x8 VopA, qs16x8 VopB, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00254">fixed_point.h:254</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a287e2fc366c312b468382c95bb90f91f"><div class="ttname"><a href="helpers_8h.xhtml#a287e2fc366c312b468382c95bb90f91f">VLOAD</a></div><div class="ttdeci">#define VLOAD(size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00037">helpers.h:37</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a527bfdf5eeb306f1cf01c4a8e29f38e0"><div class="ttname"><a href="helpers_8h.xhtml#a527bfdf5eeb306f1cf01c4a8e29f38e0">CONVERT_TO_VECTOR_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_VECTOR_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00090">helpers.h:90</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_aa66a7b6b0420e54ec173743e6c5bfc45"><div class="ttname"><a href="gemm_8cl.xhtml#aa66a7b6b0420e54ec173743e6c5bfc45">gemm_interleave4x4_8bit</a></div><div class="ttdeci">__kernel void gemm_interleave4x4_8bit(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel reshapes the input matrix transposing each 4x4 block and interleaving the values...</div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00223">gemm.cl:223</a></div></div> |
| <div class="ttc" id="convolution3x3_8cl_xhtml_afb8c72ce35c4a1f4a2588d6573e54aa1"><div class="ttname"><a href="convolution3x3_8cl.xhtml#afb8c72ce35c4a1f4a2588d6573e54aa1">DATA_TYPE</a></div><div class="ttdeci">#define DATA_TYPE</div><div class="ttdef"><b>Definition:</b> <a href="convolution3x3_8cl_source.xhtml#l00027">convolution3x3.cl:27</a></div></div> |
| <div class="ttc" id="namespacearm__compute_xhtml_a73e2825fd61d349c5ca2f5313e3c8ea1"><div class="ttname"><a href="namespacearm__compute.xhtml#a73e2825fd61d349c5ca2f5313e3c8ea1">arm_compute::half</a></div><div class="ttdeci">half_float::half half</div><div class="ttdoc">16-bit floating point type </div><div class="ttdef"><b>Definition:</b> <a href="arm__compute_2core_2_types_8h_source.xhtml#l00039">Types.h:39</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a22f42fcf2077d951271df83b55c1a71a"><div class="ttname"><a href="helpers_8h.xhtml#a22f42fcf2077d951271df83b55c1a71a">IMAGE_DECLARATION</a></div><div class="ttdeci">#define IMAGE_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00060">helpers.h:60</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_ad59f32bf4f61044a698ff4da02080e5a"><div class="ttname"><a href="fixed__point_8h.xhtml#ad59f32bf4f61044a698ff4da02080e5a">mla_sat_qs8x16</a></div><div class="ttdeci">qs8x16 mla_sat_qs8x16(qs8x16 VopA, qs8x16 VopB, qs8x16 VopC, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00275">fixed_point.h:275</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a009469e4d9b8fce3b6d5e97d2077827d"><div class="ttname"><a href="helpers_8h.xhtml#a009469e4d9b8fce3b6d5e97d2077827d">offset</a></div><div class="ttdeci">__global uchar * offset(const Image *img, int x, int y)</div><div class="ttdoc">Get the pointer position of a Image. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00295">helpers.h:295</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_acb282042d1edeeaa3cc979a206f78b54"><div class="ttname"><a href="helpers_8h.xhtml#acb282042d1edeeaa3cc979a206f78b54">VSTORE</a></div><div class="ttdeci">#define VSTORE(size)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00040">helpers.h:40</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a7a3c295a894e0529251bd4529b36a574"><div class="ttname"><a href="fixed__point_8h.xhtml#a7a3c295a894e0529251bd4529b36a574">mla_sat_qs16x8</a></div><div class="ttdeci">qs16x8 mla_sat_qs16x8(qs16x8 VopA, qs16x8 VopB, qs16x8 VopC, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00276">fixed_point.h:276</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml"><div class="ttname"><a href="helpers_8h.xhtml">helpers.h</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_aebe814363556c244be043b13e7969197"><div class="ttname"><a href="helpers_8h.xhtml#aebe814363556c244be043b13e7969197">CONVERT_TO_IMAGE_STRUCT</a></div><div class="ttdeci">#define CONVERT_TO_IMAGE_STRUCT(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00096">helpers.h:96</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a50b6bfd3a44a0ac605e88135e230d1e5"><div class="ttname"><a href="fixed__point_8h.xhtml#a50b6bfd3a44a0ac605e88135e230d1e5">mlal_sat_qs16x8</a></div><div class="ttdeci">qs32x8 mlal_sat_qs16x8(qs32x8 VopA, qs16x8 VopB, qs16x8 VopC, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00296">fixed_point.h:296</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml"><div class="ttname"><a href="fixed__point_8h.xhtml">fixed_point.h</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a40a6eb9f2a7712f08d6bb8ff6c9e6ca7"><div class="ttname"><a href="helpers_8h.xhtml#a40a6eb9f2a7712f08d6bb8ff6c9e6ca7">VECTOR_DECLARATION</a></div><div class="ttdeci">#define VECTOR_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00054">helpers.h:54</a></div></div> |
| <div class="ttc" id="struct_image_xhtml"><div class="ttname"><a href="struct_image.xhtml">Image</a></div><div class="ttdoc">Structure to hold Image information. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00134">helpers.h:134</a></div></div> |
| <div class="ttc" id="helpers_8h_xhtml_a6b83038822d1ae7ab619b684ed3b7fc0"><div class="ttname"><a href="helpers_8h.xhtml#a6b83038822d1ae7ab619b684ed3b7fc0">TENSOR3D_DECLARATION</a></div><div class="ttdeci">#define TENSOR3D_DECLARATION(name)</div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00068">helpers.h:68</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_a75c35ac120f123e9cce56c43ba3f48d2"><div class="ttname"><a href="gemm_8cl.xhtml#a75c35ac120f123e9cce56c43ba3f48d2">gemm_transpose1x8</a></div><div class="ttdeci">__kernel void gemm_transpose1x8(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel computes the "vector" 1x8 transposition of input matrix. </div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00077">gemm.cl:77</a></div></div> |
| <div class="ttc" id="struct_image_xhtml_acf52c23cbd7424606c10a606524e3e32"><div class="ttname"><a href="struct_image.xhtml#acf52c23cbd7424606c10a606524e3e32">Image::ptr</a></div><div class="ttdeci">__global uchar * ptr</div><div class="ttdoc">Pointer to the starting postion of the buffer. </div><div class="ttdef"><b>Definition:</b> <a href="helpers_8h_source.xhtml#l00136">helpers.h:136</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_a560b2d50aa886edc0f5daf4fe729717f"><div class="ttname"><a href="gemm_8cl.xhtml#a560b2d50aa886edc0f5daf4fe729717f">gemm_interleave4x4_32bit</a></div><div class="ttdeci">__kernel void gemm_interleave4x4_32bit(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel reshapes the input matrix transposing each 4x4 block and interleaving the values...</div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00141">gemm.cl:141</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_ac5eba21b9ac015e1454044aeba93546c"><div class="ttname"><a href="gemm_8cl.xhtml#ac5eba21b9ac015e1454044aeba93546c">gemm_transpose1x4</a></div><div class="ttdeci">__kernel void gemm_transpose1x4(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel computes the "vector" 1x4 transposition of input matrix. </div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00045">gemm.cl:45</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a36f754c05b6fddf6df0d8d0a74f8159f"><div class="ttname"><a href="fixed__point_8h.xhtml#a36f754c05b6fddf6df0d8d0a74f8159f">VEC_DATA_TYPE</a></div><div class="ttdeci">#define VEC_DATA_TYPE(type, size)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00093">fixed_point.h:93</a></div></div> |
| <div class="ttc" id="gemm_8cl_xhtml_a2f32d740e780059f68da8aa589ed0a5b"><div class="ttname"><a href="gemm_8cl.xhtml#a2f32d740e780059f68da8aa589ed0a5b">gemm_interleave4x4_16bit</a></div><div class="ttdeci">__kernel void gemm_interleave4x4_16bit(__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes)</div><div class="ttdoc">This OpenCL kernel reshapes the input matrix transposing each 4x4 block and interleaving the values...</div><div class="ttdef"><b>Definition:</b> <a href="gemm_8cl_source.xhtml#l00182">gemm.cl:182</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a09f5d764b62b47a1894a4cb71a0a62f6"><div class="ttname"><a href="fixed__point_8h.xhtml#a09f5d764b62b47a1894a4cb71a0a62f6">mlal_sat_qs8x8</a></div><div class="ttdeci">qs16x8 mlal_sat_qs8x8(qs16x8 VopA, qs8x8 VopB, qs8x8 VopC, int fixed_point_position)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00295">fixed_point.h:295</a></div></div> |
| <div class="ttc" id="fixed__point_8h_xhtml_a6b1acbaff6cb3bad4edda9a93dac1f9e"><div class="ttname"><a href="fixed__point_8h.xhtml#a6b1acbaff6cb3bad4edda9a93dac1f9e">ADD_SAT_OP_EXPAND</a></div><div class="ttdeci">#define ADD_SAT_OP_EXPAND(a, b, type, size)</div><div class="ttdef"><b>Definition:</b> <a href="fixed__point_8h_source.xhtml#l00178">fixed_point.h:178</a></div></div> |
| </div><!-- fragment --></div><!-- contents --> |
| </div><!-- doc-content --> |
| <!-- start footer part --> |
| <div id="nav-path" class="navpath"><!-- id is needed for treeview function! --> |
| <ul> |
| <li class="navelem"><a class="el" href="dir_55b5a5006e943fb664ff8cff3cfe7768.xhtml">src</a></li><li class="navelem"><a class="el" href="dir_f6c3ae42d7e27145d0a3c3e38ca1c000.xhtml">core</a></li><li class="navelem"><a class="el" href="dir_1b7ae9123c88e650426d50d95c19d414.xhtml">CL</a></li><li class="navelem"><a class="el" href="dir_b7c3d80b0dadb2139bc73b7161751c12.xhtml">cl_kernels</a></li><li class="navelem"><a class="el" href="gemm_8cl.xhtml">gemm.cl</a></li> |
| <li class="footer">Generated on Thu Oct 12 2017 14:26:33 for Compute Library by |
| <a href="http://www.doxygen.org/index.html"> |
| <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.6 </li> |
| </ul> |
| </div> |
| </body> |
| </html> |
