Google Git
Sign in
apache / mxnet-site / d5c5a235eef72f64ccaf224b834a20ef3287469b / . / versions / 1.9.1 / api / cpp / docs / api / dot__engine-inl_8h_source.html
blob: 49707082a7b5673cba4e009afeb917c81899fd5e [file] [log] [blame]
<!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 1.8.13"/>
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<title>mxnet: /work/mxnet/3rdparty/mshadow/mshadow/dot_engine-inl.h 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="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/searchdata.js"></script>
<script type="text/javascript" src="search/search.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 id="projectalign" style="padding-left: 0.5em;">
<div id="projectname">mxnet
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.13 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<script type="text/javascript" src="menudata.js"></script>
<script type="text/javascript" src="menu.js"></script>
<script type="text/javascript">
$(function() {
initMenu('',true,false,'search.php','Search');
$(document).ready(function() { init_search(); });
});
</script>
<div id="main-nav"></div>
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
</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 id="nav-path" class="navpath">
<ul>
<li class="navelem"><a class="el" href="dir_8cab8f464681f7cc51cee77e79a434cd.html">3rdparty</a></li><li class="navelem"><a class="el" href="dir_3e48ced36faa4eaa1b41f6d960bf0edb.html">mshadow</a></li><li class="navelem"><a class="el" href="dir_00b035bb2ad81894e6ad291054ea5f82.html">mshadow</a></li> </ul>
</div>
</div><!-- top -->
<div class="header">
<div class="headertitle">
<div class="title">dot_engine-inl.h</div> </div>
</div><!--header-->
<div class="contents">
<a href="dot__engine-inl_8h.html">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno"> 1</span>&#160;<span class="comment">/*</span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span>&#160;<span class="comment"> * Licensed to the Apache Software Foundation (ASF) under one</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span>&#160;<span class="comment"> * or more contributor license agreements. See the NOTICE file</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span>&#160;<span class="comment"> * distributed with this work for additional information</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span>&#160;<span class="comment"> * regarding copyright ownership. The ASF licenses this file</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span>&#160;<span class="comment"> * to you under the Apache License, Version 2.0 (the</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span>&#160;<span class="comment"> * &quot;License&quot;); you may not use this file except in compliance</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span>&#160;<span class="comment"> * with the License. You may obtain a copy of the License at</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span>&#160;<span class="comment"> * http://www.apache.org/licenses/LICENSE-2.0</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span>&#160;<span class="comment"> * Unless required by applicable law or agreed to in writing,</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span>&#160;<span class="comment"> * software distributed under the License is distributed on an</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span>&#160;<span class="comment"> * &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span>&#160;<span class="comment"> * KIND, either express or implied. See the License for the</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span>&#160;<span class="comment"> * specific language governing permissions and limitations</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span>&#160;<span class="comment"> * under the License.</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span>&#160;</div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span>&#160;<span class="preprocessor">#ifndef MSHADOW_DOT_ENGINE_INL_H_</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span>&#160;<span class="preprocessor">#define MSHADOW_DOT_ENGINE_INL_H_</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span>&#160;</div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span>&#160;<span class="preprocessor">#include &lt;vector&gt;</span></div><div class="line"><a name="l00029"></a><span class="lineno"> 29</span>&#160;<span class="preprocessor">#include &quot;<a class="code" href="3rdparty_2mshadow_2mshadow_2base_8h.html">./base.h</a>&quot;</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span>&#160;<span class="preprocessor">#include &quot;<a class="code" href="implicit__gemm_8h.html">./extension/implicit_gemm.h</a>&quot;</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span>&#160;</div><div class="line"><a name="l00032"></a><span class="lineno"> 32</span>&#160;<span class="preprocessor">#ifdef __CUDACC__</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span>&#160;<span class="preprocessor">#include &quot;./cuda/tensor_gpu-inl.cuh&quot;</span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span>&#160;<span class="preprocessor">#endif // #ifdef __CUDACC__</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span>&#160;</div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span>&#160;<span class="keyword">namespace </span><a class="code" href="namespacemshadow.html">mshadow</a> {</div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> Device, <span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span>&#160;<span class="keyword">inline</span> <span class="keywordtype">void</span> <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(DType **dst, DType *src, <span class="keywordtype">int</span> num, <span class="keywordtype">int</span> stride,</div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span>&#160; Stream&lt;Device&gt; *stream);</div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00049"></a><span class="lineno"><a class="line" href="namespacemshadow.html#a9aefcfd5ca44f5e146d2274d64acc239"> 49</a></span>&#160;<span class="keyword">inline</span> <span class="keywordtype">void</span> <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(DType **dst, DType *src, <span class="keywordtype">int</span> num, <span class="keywordtype">int</span> stride,</div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>&#160; <a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream) {</div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; num; i++) {</div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>&#160; dst[i] = src + i * stride;</div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>&#160; }</div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>&#160;}</div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span>&#160;<span class="preprocessor">#ifdef __CUDACC__</span></div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>&#160;<span class="keyword">namespace </span>cuda {};</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>&#160;<span class="keyword">inline</span> <span class="keywordtype">void</span> <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(DType **dst, DType *src, <span class="keywordtype">int</span> num, <span class="keywordtype">int</span> stride,</div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span>&#160; <a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream) {</div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span>&#160; <a class="code" href="namespacemshadow.html#a9aefcfd5ca44f5e146d2274d64acc239">cuda::GetBatchedView</a>(dst, src, num, stride, stream);</div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span>&#160;}</div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span>&#160;<span class="preprocessor">#endif // #ifdef __CUDACC__</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span>&#160;</div><div class="line"><a name="l00064"></a><span class="lineno"><a class="line" href="namespacemshadow_1_1expr.html"> 64</a></span>&#160;<span class="keyword">namespace </span>expr {</div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span>&#160;<span class="comment">//---------------------------------------------------------------------</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>&#160;<span class="comment">// Matrix Multiplications, depends on BLAS Engine</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>&#160;<span class="comment">//---------------------------------------------------------------------</span></div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> SV, <span class="keyword">typename</span> Device, <span class="keywordtype">int</span> ddim, <span class="keywordtype">int</span> ldim,</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>&#160; <span class="keywordtype">int</span> rdim, <span class="keywordtype">bool</span> ltrans, <span class="keywordtype">bool</span> rtrans, <span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00070"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine.html"> 70</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1DotEngine.html">DotEngine</a> {</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> Eval(<a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;Device, ddim, DType&gt;</a> *p_dst,</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;Device, ldim, DType&gt;</a> &amp;lhs,</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;Device, rdim, DType&gt;</a> &amp;rhs,</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>&#160; DType scale);</div><div class="line"><a name="l00075"></a><span class="lineno"> 75</span>&#160;};</div><div class="line"><a name="l00076"></a><span class="lineno"> 76</span>&#160;<span class="comment">// handles the dot, use CblasColMajor</span></div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> Device, <span class="keyword">typename</span> DType = default_real_t&gt;</div><div class="line"><a name="l00078"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html"> 78</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a> {</div><div class="line"><a name="l00079"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#ac1ee5340c4bf6e123db07dfc57a952d7"> 79</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">bool</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#ac1ee5340c4bf6e123db07dfc57a952d7">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>&#160; <span class="keywordflow">return</span> t ? true : <span class="keyword">false</span>;</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>&#160; }</div><div class="line"><a name="l00082"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#ac0b67fda3ecdf2087b816602ad43eda3"> 82</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#ac0b67fda3ecdf2087b816602ad43eda3">SetStream</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream) {</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>&#160; }</div><div class="line"><a name="l00084"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#a66dd260062b01a60d54370cb71871830"> 84</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#a66dd260062b01a60d54370cb71871830">gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, DType alpha,</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>&#160; <span class="keyword">const</span> DType *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> DType *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>&#160; DType beta, DType *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>&#160; }</div><div class="line"><a name="l00091"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#a85dfe7b9fe210cb4dbe06a97fa81ecda"> 91</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#a85dfe7b9fe210cb4dbe06a97fa81ecda">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, DType alpha,</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>&#160; <span class="keyword">const</span> DType *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> DType *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>&#160; DType beta, DType *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>&#160; DType **workspace) {</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span>&#160; }</div><div class="line"><a name="l00099"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#a6186eb5d53e0a9d956675550b5cdb41a"> 99</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#a6186eb5d53e0a9d956675550b5cdb41a">gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>&#160; DType alpha, <span class="keyword">const</span> DType *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>&#160; <span class="keyword">const</span> DType *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>&#160; DType beta, DType *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00105"></a><span class="lineno"> 105</span>&#160; }</div><div class="line"><a name="l00106"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#aea1017183bb11680f4f041c13c5d3a7f"> 106</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#aea1017183bb11680f4f041c13c5d3a7f">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>&#160; DType alpha, <span class="keyword">const</span> DType *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>&#160; <span class="keyword">const</span> DType *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>&#160; DType beta, DType *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>&#160; }</div><div class="line"><a name="l00113"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#a3785a79812f5e01713845480a790601c"> 113</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#a3785a79812f5e01713845480a790601c">ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, DType alpha,</div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>&#160; <span class="keyword">const</span> DType *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>&#160; <span class="keyword">const</span> DType *Y, <span class="keywordtype">int</span> incY, DType *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>&#160; }</div><div class="line"><a name="l00119"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#abd8987afc8ece12882f139f39c8fca7a"> 119</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#abd8987afc8ece12882f139f39c8fca7a">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, DType alpha,</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>&#160; <span class="keyword">const</span> DType *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>&#160; <span class="keyword">const</span> DType *Y, <span class="keywordtype">int</span> incY, DType *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00124"></a><span class="lineno"> 124</span>&#160; }</div><div class="line"><a name="l00125"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine.html#af21b0de1c3d687671e36cffb464d5988"> 125</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html#af21b0de1c3d687671e36cffb464d5988">dot</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;Device&gt;</a> *stream,</div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>&#160; <span class="keyword">const</span> DType* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>&#160; <span class="keyword">const</span> DType* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>&#160; DType* ret) {</div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>&#160; }</div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span>&#160;};</div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>&#160;</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>&#160;<span class="preprocessor">#if MSHADOW_STAND_ALONE</span></div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;cpu, float&gt; {</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">bool</span> GetT(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>&#160; <span class="keywordflow">return</span> t ? true : <span class="keyword">false</span>;</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>&#160; }</div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> SetStream(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream) {</div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>&#160; }</div><div class="line"><a name="l00142"></a><span class="lineno"> 142</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> gemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>&#160; <span class="keywordflow">if</span> (alpha == 1.0f &amp;&amp; beta == 0.0f) {</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>&#160; <span class="keywordtype">bool</span> transpose_left = transb;</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>&#160; <span class="keywordtype">bool</span> transpose_right = transa;</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, float&gt;</a> lhs((<span class="keywordtype">float</span>*)B, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(transpose_left ? k : n, transpose_left ? n : k)); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, float&gt;</a> rhs((<span class="keywordtype">float</span>*)A, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(transpose_right ? m : k, transpose_right ? k : m)); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, float&gt;</a> dst(C, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(m, n));</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>&#160; <span class="keywordflow">if</span> (!transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (!transpose_left &amp;&amp; transpose_right) {</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>()); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>(), rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>&#160; } <span class="keywordflow">else</span> {</div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>&#160; }</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>&#160; } <span class="keywordflow">else</span> {</div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>&#160; }</div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>&#160; }</div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_gemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>&#160; <span class="keywordtype">float</span> **workspace) {</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>&#160; }</div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>&#160; }</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> gemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>&#160; <span class="keywordtype">float</span> alpha, <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>&#160; }</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_gemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>&#160; <span class="keywordtype">float</span> alpha, <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>&#160; }</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> ger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>&#160; }</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_ger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00203"></a><span class="lineno"> 203</span>&#160; }</div><div class="line"><a name="l00204"></a><span class="lineno"> 204</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="namespacemshadow_1_1expr.html#a496d2f5fb874d0a8d59b6a0b4374e2df">dot</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>&#160; <span class="keywordtype">float</span>* ret) {</div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>&#160; }</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>&#160;};</div><div class="line"><a name="l00212"></a><span class="lineno"> 212</span>&#160;</div><div class="line"><a name="l00213"></a><span class="lineno"> 213</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;cpu, double&gt; {</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">bool</span> GetT(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>&#160; <span class="keywordflow">return</span> t ? true : <span class="keyword">false</span>;</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>&#160; }</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> SetStream(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream) {</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>&#160; }</div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> gemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>&#160; <span class="keywordflow">if</span> (alpha == 1.0f &amp;&amp; beta == 0.0f) {</div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>&#160; <span class="keywordtype">bool</span> transpose_left = transb;</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>&#160; <span class="keywordtype">bool</span> transpose_right = transa;</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, double&gt;</a> lhs((<span class="keywordtype">double</span>*)B, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(transpose_left ? k : n, transpose_left ? n : k)); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, double&gt;</a> rhs((<span class="keywordtype">double</span>*)A, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(transpose_right ? m : k, transpose_right ? k : m)); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;cpu, 2, double&gt;</a> dst(C, <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(m, n));</div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>&#160; <span class="keywordflow">if</span> (!transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (!transpose_left &amp;&amp; transpose_right) {</div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>()); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>(), rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>&#160; } <span class="keywordflow">else</span> {</div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>&#160; }</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>&#160; } <span class="keywordflow">else</span> {</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>&#160; }</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>&#160; }</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_gemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>&#160; <span class="keywordtype">double</span> **workspace) {</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>&#160; }</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>&#160; }</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> gemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>&#160; <span class="keywordtype">double</span> alpha, <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>&#160; }</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_gemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>&#160; <span class="keywordtype">double</span> alpha, <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span>&#160; }</div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> ger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>&#160; }</div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> batched_ger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00281"></a><span class="lineno"> 281</span>&#160; }</div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="namespacemshadow_1_1expr.html#a496d2f5fb874d0a8d59b6a0b4374e2df">dot</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>&#160; <span class="keywordtype">double</span>* ret) {</div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>&#160; }</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>&#160;};</div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span>&#160;</div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>&#160;<span class="preprocessor">#elif (MSHADOW_USE_MKL || MSHADOW_USE_CBLAS) // NOLINT(*)</span></div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00293"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html"> 293</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;cpu, float&gt; {</div><div class="line"><a name="l00294"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a179fabfc4904fe6ba064116dd27be8d7"> 294</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> CBLAS_TRANSPOSE <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a179fabfc4904fe6ba064116dd27be8d7">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>&#160; <span class="keywordflow">return</span> t ? CblasTrans : CblasNoTrans;</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>&#160; }</div><div class="line"><a name="l00297"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#abdf3b3f658b9d8277c1351229ad3e6ce"> 297</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#abdf3b3f658b9d8277c1351229ad3e6ce">SetStream</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream) {</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>&#160; }</div><div class="line"><a name="l00299"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ad3f8beddfbc19c0a25be5f856a540f39"> 299</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ad3f8beddfbc19c0a25be5f856a540f39">gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>&#160; cblas_sgemm(CblasColMajor, GetT(transa), GetT(transb),</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>&#160; m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>&#160; }</div><div class="line"><a name="l00307"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a9977ea76d73267cf26680ec0b34ab062"> 307</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a9977ea76d73267cf26680ec0b34ab062">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>&#160; <span class="keywordtype">float</span> **workspace) {</div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>&#160;<span class="preprocessor">#if (MSHADOW_USE_MKL &amp;&amp; INTEL_MKL_VERSION &gt;= 20160000)</span></div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>&#160; <span class="comment">// since same m/n/k is used for all single gemms, so we put all gemms into one group</span></div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> GROUP_SIZE = 1;</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>&#160; MKL_INT p_m[GROUP_SIZE] = {m};</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>&#160; MKL_INT p_n[GROUP_SIZE] = {n};</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>&#160; MKL_INT p_k[GROUP_SIZE] = {k};</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>&#160; MKL_INT p_lda[GROUP_SIZE] = {lda};</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>&#160; MKL_INT p_ldb[GROUP_SIZE] = {ldb};</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>&#160; MKL_INT p_ldc[GROUP_SIZE] = {ldc};</div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span>&#160;</div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span>&#160; <span class="keywordtype">float</span> p_alpha[GROUP_SIZE] = {alpha};</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>&#160; <span class="keywordtype">float</span> p_beta[GROUP_SIZE] = {beta};</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>&#160;</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>&#160; CBLAS_TRANSPOSE cblas_a_trans = GetT(transa);</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>&#160; CBLAS_TRANSPOSE cblas_b_trans = GetT(transb);</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>&#160;</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>&#160; MKL_INT p_group_sizeb[GROUP_SIZE] = {batch_count};</div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>&#160; CBLAS_TRANSPOSE p_transa[GROUP_SIZE] = {cblas_a_trans};</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>&#160; CBLAS_TRANSPOSE p_transb[GROUP_SIZE] = {cblas_b_trans};</div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>&#160;</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>&#160; std::vector&lt;const float*&gt; pp_A(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>&#160; std::vector&lt;const float*&gt; pp_B(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>&#160; std::vector&lt;float*&gt; pp_C(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>&#160;</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>&#160; <span class="keyword">auto</span> m_k = m * k;</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>&#160; <span class="keyword">auto</span> k_n = k * n;</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>&#160; <span class="keyword">auto</span> m_n = m * n;</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>&#160;</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; i++) {</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>&#160; pp_A[i] = A + i * m_k;</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>&#160; pp_B[i] = B + i * k_n;</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>&#160; pp_C[i] = C + i * m_n;</div><div class="line"><a name="l00345"></a><span class="lineno"> 345</span>&#160; }</div><div class="line"><a name="l00346"></a><span class="lineno"> 346</span>&#160;</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>&#160; cblas_sgemm_batch(CblasColMajor, p_transa, p_transb,</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>&#160; p_m, p_n, p_k, p_alpha, pp_A.data(), p_lda, pp_B.data(),</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>&#160; p_ldb, p_beta, pp_C.data(), p_ldc, GROUP_SIZE, p_group_sizeb);</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>&#160; }</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>&#160; }</div><div class="line"><a name="l00358"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5c9f30798d4e51a1f989621063e2e0ad"> 358</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5c9f30798d4e51a1f989621063e2e0ad">gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>&#160; <span class="keywordtype">float</span> alpha, <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>&#160; cblas_sgemv(CblasColMajor, GetT(trans), m, n, alpha,</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>&#160; A, lda, X, incX, beta, Y, incY);</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>&#160; }</div><div class="line"><a name="l00366"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ac38c4c955ccbf0749acf51afcb5143d4"> 366</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ac38c4c955ccbf0749acf51afcb5143d4">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>&#160; <span class="keywordtype">float</span> alpha, <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>&#160; gemv(stream, trans, m, n, alpha, A + i * m * n, lda,</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>&#160; X + i * (trans ? m : n) * incX, incX,</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>&#160; beta, Y + i * (trans ? n : m) * incY, incY);</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>&#160; }</div><div class="line"><a name="l00376"></a><span class="lineno"> 376</span>&#160; }</div><div class="line"><a name="l00377"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a3d822069b1c05de013ff06e4680c6454"> 377</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a3d822069b1c05de013ff06e4680c6454">ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>&#160; cblas_sger(CblasColMajor, m, n, alpha, X, incX, Y, incY, A, lda);</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>&#160; }</div><div class="line"><a name="l00383"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a8895cfca8053511f48053a093d43a7f6"> 383</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a8895cfca8053511f48053a093d43a7f6">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>&#160; ger(stream, m, n, alpha, X + i * m * incX, incX, Y + i * n * incY, incY,</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>&#160; A + i * lda * n, lda);</div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>&#160; }</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>&#160; }</div><div class="line"><a name="l00392"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5bf25e6bdccbe70d1f986701ce6414a2"> 392</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5bf25e6bdccbe70d1f986701ce6414a2">dot</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>&#160; <span class="keywordtype">float</span>* ret) {</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>&#160; *ret = cblas_sdot(n, X, incX, Y, incY);</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span>&#160; }</div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span>&#160;};</div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>&#160;</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00402"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html"> 402</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;cpu, double&gt; {</div><div class="line"><a name="l00403"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a470e53711519b24cad68bbd41d600052"> 403</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> CBLAS_TRANSPOSE <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a470e53711519b24cad68bbd41d600052">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>&#160; <span class="keywordflow">return</span> t ? CblasTrans : CblasNoTrans;</div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span>&#160; }</div><div class="line"><a name="l00406"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ad88269ee104f5310e0a5ccc6b1fb1339"> 406</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ad88269ee104f5310e0a5ccc6b1fb1339">SetStream</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream) {</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>&#160; }</div><div class="line"><a name="l00408"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a19e843390ec50c3064698408603fc63a"> 408</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a19e843390ec50c3064698408603fc63a">gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>&#160; cblas_dgemm(CblasColMajor, GetT(transa), GetT(transb),</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>&#160; m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>&#160; }</div><div class="line"><a name="l00416"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a621f673c37d15f9e1acb76bd7be3be82"> 416</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a621f673c37d15f9e1acb76bd7be3be82">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>&#160; <span class="keywordtype">double</span> **workspace) {</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>&#160;<span class="preprocessor">#if (MSHADOW_USE_MKL &amp;&amp; INTEL_MKL_VERSION &gt;= 20160000)</span></div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>&#160; <span class="comment">// since same m/n/k is used for all single gemms, so we put all gemms into one group</span></div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>&#160; <span class="keyword">const</span> <span class="keywordtype">int</span> GROUP_SIZE = 1;</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>&#160; MKL_INT p_m[GROUP_SIZE] = {m};</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>&#160; MKL_INT p_n[GROUP_SIZE] = {n};</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>&#160; MKL_INT p_k[GROUP_SIZE] = {k};</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>&#160; MKL_INT p_lda[GROUP_SIZE] = {lda};</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>&#160; MKL_INT p_ldb[GROUP_SIZE] = {ldb};</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>&#160; MKL_INT p_ldc[GROUP_SIZE] = {ldc};</div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>&#160;</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span>&#160; <span class="keywordtype">double</span> p_alpha[GROUP_SIZE] = {alpha};</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>&#160; <span class="keywordtype">double</span> p_beta[GROUP_SIZE] = {beta};</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>&#160;</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>&#160; CBLAS_TRANSPOSE cblas_a_trans = GetT(transa);</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>&#160; CBLAS_TRANSPOSE cblas_b_trans = GetT(transb);</div><div class="line"><a name="l00437"></a><span class="lineno"> 437</span>&#160;</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span>&#160; MKL_INT p_group_sizeb[GROUP_SIZE] = {batch_count};</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>&#160; CBLAS_TRANSPOSE p_transa[GROUP_SIZE] = {cblas_a_trans};</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>&#160; CBLAS_TRANSPOSE p_transb[GROUP_SIZE] = {cblas_b_trans};</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>&#160;</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>&#160; std::vector&lt;const double*&gt; pp_A(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>&#160; std::vector&lt;const double*&gt; pp_B(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>&#160; std::vector&lt;double*&gt; pp_C(batch_count, <span class="keyword">nullptr</span>);</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>&#160;</div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>&#160; <span class="keyword">auto</span> m_k = m * k;</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span>&#160; <span class="keyword">auto</span> k_n = k * n;</div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span>&#160; <span class="keyword">auto</span> m_n = m * n;</div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span>&#160;</div><div class="line"><a name="l00450"></a><span class="lineno"> 450</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; i++) {</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>&#160; pp_A[i] = A + i * m_k;</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>&#160; pp_B[i] = B + i * k_n;</div><div class="line"><a name="l00453"></a><span class="lineno"> 453</span>&#160; pp_C[i] = C + i * m_n;</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span>&#160; }</div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span>&#160;</div><div class="line"><a name="l00456"></a><span class="lineno"> 456</span>&#160; cblas_dgemm_batch(CblasColMajor, p_transa, p_transb,</div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>&#160; p_m, p_n, p_k, p_alpha, pp_A.data(), p_lda, pp_B.data(),</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>&#160; p_ldb, p_beta, pp_C.data(), p_ldc, GROUP_SIZE, p_group_sizeb);</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>&#160; }</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>&#160; }</div><div class="line"><a name="l00467"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aefac67f858a67b1bba74bc06ed2c3e99"> 467</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aefac67f858a67b1bba74bc06ed2c3e99">gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span>&#160; cblas_dgemv(CblasColMajor, GetT(trans), m, n, alpha,</div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span>&#160; A, lda, X, incX, beta, Y, incY);</div><div class="line"><a name="l00474"></a><span class="lineno"> 474</span>&#160; }</div><div class="line"><a name="l00475"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa8c29bbc3c224b7da63a09d1f3d7727e"> 475</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa8c29bbc3c224b7da63a09d1f3d7727e">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00477"></a><span class="lineno"> 477</span>&#160; <span class="keywordtype">double</span> alpha, <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>&#160; gemv(stream, trans, m, n, alpha, A + i * m * n, lda,</div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>&#160; X + i * (trans ? m : n) * incX, incX,</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span>&#160; beta, Y + i * (trans ? n : m) * incY, incY);</div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>&#160; }</div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>&#160; }</div><div class="line"><a name="l00486"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ab923458f56a77bc3e68d541f85ce8599"> 486</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ab923458f56a77bc3e68d541f85ce8599">ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00489"></a><span class="lineno"> 489</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span>&#160; cblas_dger(CblasColMajor, m, n, alpha, X, incX, Y, incY, A, lda);</div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span>&#160; }</div><div class="line"><a name="l00492"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a8a5491682d961329ceed384c6f8c9931"> 492</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a8a5491682d961329ceed384c6f8c9931">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00493"></a><span class="lineno"> 493</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00495"></a><span class="lineno"> 495</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00496"></a><span class="lineno"> 496</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span>&#160; ger(stream, m, n, alpha, X + i * m * incX, incX, Y + i * n * incY, incY,</div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span>&#160; A + i * lda * n, lda);</div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span>&#160; }</div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>&#160; }</div><div class="line"><a name="l00501"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa6492ff9a5684ad19cf90b875e00b365"> 501</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa6492ff9a5684ad19cf90b875e00b365">dot</a>(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;cpu&gt;</a> *stream,</div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>&#160; <span class="keywordtype">double</span>* ret) {</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>&#160; *ret = cblas_ddot(n, X, incX, Y, incY);</div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>&#160; }</div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>&#160;};</div><div class="line"><a name="l00509"></a><span class="lineno"> 509</span>&#160;<span class="preprocessor">#endif // MSHADOW_USE_CBLAS || MSHADOW_USE_MKL || MSHADOW_STAND_ALONE</span></div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span>&#160;<span class="comment">// CuBLAS redirect code</span></div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span>&#160;<span class="preprocessor">#if MSHADOW_USE_CUDA</span></div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>&#160;<span class="comment">// All CuBLAS goes to here, use legacy API: not threadsafe</span></div><div class="line"><a name="l00513"></a><span class="lineno"> 513</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00514"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html"> 514</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;<a class="code" href="structmshadow_1_1gpu.html">gpu</a>, half::half_t&gt; {</div><div class="line"><a name="l00515"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ac742f67973dd9ce1c444f0c476749d10"> 515</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> cublasOperation_t <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ac742f67973dd9ce1c444f0c476749d10">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span>&#160; <span class="keywordflow">return</span> t ? CUBLAS_OP_T : CUBLAS_OP_N;</div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>&#160; }</div><div class="line"><a name="l00518"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a2091ee54add7a9ff95c6e1f08bed21d5"> 518</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a2091ee54add7a9ff95c6e1f08bed21d5">SetStream</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream) {</div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>&#160; cublasStatus_t err = cublasSetStream(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>&#160; <a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetStream</a>(stream));</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas set stream fail&quot;</span>;</div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span>&#160; }</div><div class="line"><a name="l00523"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a8d6d8e682966822bb85b6ff6fee32ce2"> 523</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a8d6d8e682966822bb85b6ff6fee32ce2">gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, half::half_t alpha,</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>&#160; <span class="keyword">const</span> half::half_t *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span>&#160; <span class="keyword">const</span> half::half_t *B, <span class="keywordtype">int</span> ldb, half::half_t beta,</div><div class="line"><a name="l00528"></a><span class="lineno"> 528</span>&#160; half::half_t *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00529"></a><span class="lineno"> 529</span>&#160;<span class="preprocessor">#if defined(CUDA_VERSION) &amp;&amp; CUDA_VERSION &gt;= 7050</span></div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>&#160; <span class="comment">// Always use pseudo-fp16: fp32 compute with fp16 I/O.</span></div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span>&#160; <span class="keywordtype">float</span> alpha_f = float(alpha); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span>&#160; <span class="keywordtype">float</span> beta_f = float(beta); <span class="comment">// NOLINT(*)</span></div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>&#160;<span class="preprocessor"> #if CUDA_VERSION &gt;= 8000</span></div><div class="line"><a name="l00534"></a><span class="lineno"> 534</span>&#160; cublasStatus_t err = cublasSgemmEx(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha_f,</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span>&#160; A, CUDA_R_16F, lda, B, CUDA_R_16F,</div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span>&#160; ldb, &amp;beta_f, C, CUDA_R_16F, ldc);</div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas SgemmEx fail&quot;</span>;</div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span>&#160;<span class="preprocessor"> #else</span></div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span>&#160; cublasStatus_t err = cublasSgemmEx(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha_f,</div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span>&#160; A, CUBLAS_DATA_HALF, lda, B, CUBLAS_DATA_HALF,</div><div class="line"><a name="l00543"></a><span class="lineno"> 543</span>&#160; ldb, &amp;beta_f, C, CUBLAS_DATA_HALF, ldc);</div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas SgemmEx fail&quot;</span>;</div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span>&#160;<span class="preprocessor"> #endif // CUDA_VERSION &gt;= 8000</span></div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Require CUDA version &gt;= 7.5!&quot;</span>;</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span>&#160;<span class="preprocessor">#endif // defined(CUDA_VERSION) &amp;&amp; CUDA_VERSION &gt;= 7050</span></div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span>&#160; }</div><div class="line"><a name="l00550"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a517affe439b1378dff42a2efe8c21dd8"> 550</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a517affe439b1378dff42a2efe8c21dd8">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, half::half_t alpha,</div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>&#160; <span class="keyword">const</span> half::half_t *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> half::half_t *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>&#160; half::half_t beta, half::half_t *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>&#160; half::half_t **workspace) {</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>&#160;<span class="preprocessor">#if defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 9000</span></div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span>&#160; <span class="keywordtype">int</span> major = stream-&gt;<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html#a4ec551c8440da0c89eb728e753906936">prop</a>.major;</div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span>&#160; <span class="keywordtype">int</span> minor = stream-&gt;<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html#a4ec551c8440da0c89eb728e753906936">prop</a>.minor;</div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>&#160; <span class="comment">// fp16 is not supported before ARCH 53</span></div><div class="line"><a name="l00560"></a><span class="lineno"> 560</span>&#160; <span class="keywordflow">if</span> ((major &gt; 5) || (major == 5 &amp;&amp; minor &gt;= 3)) {</div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span>&#160; <span class="keyword">const</span> __half* A_h = <span class="keyword">reinterpret_cast&lt;</span><span class="keyword">const </span>__half*<span class="keyword">&gt;</span>(A);</div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span>&#160; <span class="keyword">const</span> __half* B_h = <span class="keyword">reinterpret_cast&lt;</span><span class="keyword">const </span>__half*<span class="keyword">&gt;</span>(B);</div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span>&#160; __half* alpha_h = <span class="keyword">reinterpret_cast&lt;</span>__half*<span class="keyword">&gt;</span>(&amp;alpha);</div><div class="line"><a name="l00564"></a><span class="lineno"> 564</span>&#160; __half* beta_h = <span class="keyword">reinterpret_cast&lt;</span>__half*<span class="keyword">&gt;</span>(&amp;beta);</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</span>&#160; __half* C_h = <span class="keyword">reinterpret_cast&lt;</span>__half*<span class="keyword">&gt;</span>(C);</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>&#160; cublasStatus_t err = cublasHgemmStridedBatched(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>&#160; GetT(transa), GetT(transb), m, n, k, alpha_h,</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>&#160; A_h, lda, m * k,</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>&#160; B_h, ldb, k * n,</div><div class="line"><a name="l00570"></a><span class="lineno"> 570</span>&#160; beta_h, C_h, ldc, m * n,</div><div class="line"><a name="l00571"></a><span class="lineno"> 571</span>&#160; batch_count);</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: HgemmStridedBatched fail&quot;</span>;</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>&#160; <span class="keywordflow">return</span>;</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span>&#160; }</div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00579"></a><span class="lineno"> 579</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span>&#160; }</div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span>&#160; }</div><div class="line"><a name="l00582"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ae014aad49673f34d8f4149ea4525820b"> 582</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ae014aad49673f34d8f4149ea4525820b">gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, half::half_t alpha,</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>&#160; <span class="keyword">const</span> half::half_t *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span>&#160; <span class="keyword">const</span> half::half_t *X, <span class="keywordtype">int</span> incX, half::half_t beta,</div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>&#160; half::half_t *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00587"></a><span class="lineno"> 587</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>&#160; }</div><div class="line"><a name="l00589"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a7dbffea32aea2ed9b58e8098c80cd05f"> 589</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a7dbffea32aea2ed9b58e8098c80cd05f">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</span>&#160; half::half_t alpha, <span class="keyword">const</span> half::half_t *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>&#160; <span class="keyword">const</span> half::half_t *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>&#160; half::half_t beta, half::half_t *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>&#160; }</div><div class="line"><a name="l00596"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a6d1f6461bd234ab76f380ac6da8f8f1c"> 596</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a6d1f6461bd234ab76f380ac6da8f8f1c">ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, half::half_t alpha,</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span>&#160; <span class="keyword">const</span> half::half_t *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span>&#160; <span class="keyword">const</span> half::half_t *Y, <span class="keywordtype">int</span> incY, half::half_t *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00600"></a><span class="lineno"> 600</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00601"></a><span class="lineno"> 601</span>&#160; }</div><div class="line"><a name="l00602"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#af75b98880cf90282c31c002966e2c3ca"> 602</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#af75b98880cf90282c31c002966e2c3ca">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, half::half_t alpha,</div><div class="line"><a name="l00604"></a><span class="lineno"> 604</span>&#160; <span class="keyword">const</span> half::half_t *X, <span class="keywordtype">int</span> incX, <span class="keyword">const</span> half::half_t *Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span>&#160; half::half_t *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00607"></a><span class="lineno"> 607</span>&#160; }</div><div class="line"><a name="l00608"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a88faac9a1b47990aed30b1b284154624"> 608</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a88faac9a1b47990aed30b1b284154624">dot</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00610"></a><span class="lineno"> 610</span>&#160; <span class="keyword">const</span> half::half_t* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span>&#160; <span class="keyword">const</span> half::half_t* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span>&#160; half::half_t *ret) {</div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span>&#160; LOG(FATAL) &lt;&lt; <span class="stringliteral">&quot;Not implmented!&quot;</span>;</div><div class="line"><a name="l00614"></a><span class="lineno"> 614</span>&#160; }</div><div class="line"><a name="l00615"></a><span class="lineno"> 615</span>&#160;};</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span>&#160;</div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00618"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html"> 618</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;<a class="code" href="structmshadow_1_1gpu.html">gpu</a>, float&gt; {</div><div class="line"><a name="l00619"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a1ef82a21b3debea986511176e88163f5"> 619</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> cublasOperation_t <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a1ef82a21b3debea986511176e88163f5">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00620"></a><span class="lineno"> 620</span>&#160; <span class="keywordflow">return</span> t ? CUBLAS_OP_T : CUBLAS_OP_N;</div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span>&#160; }</div><div class="line"><a name="l00622"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ab384dfeab544e586a57258b0ce3082c2"> 622</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ab384dfeab544e586a57258b0ce3082c2">SetStream</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream) {</div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>&#160; cublasStatus_t err = cublasSetStream(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span>&#160; <a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetStream</a>(stream));</div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: set stream fail&quot;</span>;</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span>&#160; }</div><div class="line"><a name="l00627"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a2beab1ca667698e87fe09b6c63e22b8f"> 627</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a2beab1ca667698e87fe09b6c63e22b8f">gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb, <span class="keywordtype">float</span> beta,</div><div class="line"><a name="l00632"></a><span class="lineno"> 632</span>&#160; <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span>&#160; cublasStatus_t err = cublasSgemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00635"></a><span class="lineno"> 635</span>&#160; A, lda, B, ldb, &amp;beta, C, ldc);</div><div class="line"><a name="l00636"></a><span class="lineno"> 636</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Sgemm fail&quot;</span>;</div><div class="line"><a name="l00637"></a><span class="lineno"> 637</span>&#160; }</div><div class="line"><a name="l00638"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a463ff27ac0c305235aee476f5d6b7608"> 638</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a463ff27ac0c305235aee476f5d6b7608">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00639"></a><span class="lineno"> 639</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">float</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>&#160; <span class="keywordtype">float</span> **workspace) {</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>&#160;<span class="preprocessor">#if defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 4010 &amp;&amp; CUDA_VERSION &lt; 8000</span></div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>&#160; <span class="comment">// Cast DType* to DType** using workspace as a buffer</span></div><div class="line"><a name="l00646"></a><span class="lineno"> 646</span>&#160; <span class="keywordtype">bool</span> alloc_workspace = <span class="keyword">false</span>;</div><div class="line"><a name="l00647"></a><span class="lineno"> 647</span>&#160; <span class="keywordflow">if</span> (workspace == NULL) {</div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>&#160; <span class="comment">// Allocate the workspace if it&#39;s NULL.</span></div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>&#160; <span class="comment">// TODO(sxjscience) Try to move the allocation inside Tensor, which is thread-safe.</span></div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>&#160; cudaMalloc(reinterpret_cast&lt;void**&gt;(&amp;workspace), 3 * batch_count * <span class="keyword">sizeof</span>(<span class="keywordtype">float</span>*));</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span>&#160; alloc_workspace = <span class="keyword">true</span>;</div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span>&#160; }</div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace, const_cast&lt;float*&gt;(A), batch_count, m * k, stream);</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace + batch_count,</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span>&#160; const_cast&lt;float*&gt;(B), batch_count, k * n, stream);</div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace + 2 * batch_count, C, batch_count, m * n, stream);</div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span>&#160; cublasStatus_t err = cublasSgemmBatched(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00659"></a><span class="lineno"> 659</span>&#160; (<span class="keyword">const</span> <span class="keywordtype">float</span>**)workspace, lda,</div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>&#160; (<span class="keyword">const</span> <span class="keywordtype">float</span>**)(workspace + batch_count), ldb,</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span>&#160; &amp;beta, workspace + 2 * batch_count, ldc, batch_count);</div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: SgemmBatched fail&quot;</span>;</div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span>&#160; <span class="keywordflow">if</span> (alloc_workspace) {</div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span>&#160; cudaFree(workspace);</div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>&#160; }</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>&#160;<span class="preprocessor">#elif defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 8000</span></div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span>&#160; cublasStatus_t err = cublasSgemmStridedBatched(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>&#160; A, lda, m * k,</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span>&#160; B, ldb, k * n,</div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span>&#160; &amp;beta, C, ldc, m * n,</div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span>&#160; batch_count);</div><div class="line"><a name="l00673"></a><span class="lineno"> 673</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: SgemmStridedBatched fail&quot;</span>;</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00679"></a><span class="lineno"> 679</span>&#160; }</div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span>&#160;<span class="preprocessor">#endif // defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 4010</span></div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>&#160; }</div><div class="line"><a name="l00682"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a7b194dedaa7a20eb61fce9d6ccff7690"> 682</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a7b194dedaa7a20eb61fce9d6ccff7690">gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00684"></a><span class="lineno"> 684</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX, <span class="keywordtype">float</span> beta,</div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span>&#160; <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00687"></a><span class="lineno"> 687</span>&#160; cublasStatus_t err = cublasSgemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span>&#160; GetT(trans), m, n, &amp;alpha, A, lda, X, incX, &amp;beta, Y, incY);</div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Sgemv fail&quot;</span>;</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span>&#160; }</div><div class="line"><a name="l00691"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a979610afce873cc7d4bac17a22a7518a"> 691</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a979610afce873cc7d4bac17a22a7518a">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span>&#160; <span class="keywordtype">float</span> alpha, <span class="keyword">const</span> <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>&#160; <span class="keywordtype">float</span> beta, <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>&#160; gemv(stream, trans, m, n, alpha, A + i * m * n, lda,</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span>&#160; X + i * (trans ? m : n) * incX, incX,</div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span>&#160; beta, Y + i * (trans ? n : m) * incY, incY);</div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span>&#160; }</div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>&#160; }</div><div class="line"><a name="l00702"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a3b5d0cfe43d79f1b8aadaca3ae0cf0cc"> 702</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a3b5d0cfe43d79f1b8aadaca3ae0cf0cc">ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00704"></a><span class="lineno"> 704</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span>&#160; cublasStatus_t err = cublasSger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span>&#160; m, n, &amp;alpha, X, incX, Y, incY, A, lda);</div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Sger fail&quot;</span>;</div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span>&#160; }</div><div class="line"><a name="l00710"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a76cdb254a4f93f1e1fbaaa335177bccc"> 710</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a76cdb254a4f93f1e1fbaaa335177bccc">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">float</span> alpha,</div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">float</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>&#160; ger(stream, m, n, alpha, X + i * m * incX, incX, Y + i * n * incY, incY,</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span>&#160; A + i * lda * n, lda);</div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>&#160; }</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span>&#160; }</div><div class="line"><a name="l00719"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ae1db170ffdc4480318bbaf17fe227b4b"> 719</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ae1db170ffdc4480318bbaf17fe227b4b">dot</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span>&#160; <span class="keyword">const</span> <span class="keywordtype">float</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span>&#160; <span class="keywordtype">float</span> *ret) {</div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span>&#160; cublasSetPointerMode(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>&#160; CUBLAS_POINTER_MODE_DEVICE);</div><div class="line"><a name="l00726"></a><span class="lineno"> 726</span>&#160; cublasStatus_t err = cublasSdot(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>&#160; n, X, incX, Y, incY, ret);</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Dot fail&quot;</span>;</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>&#160; cublasSetPointerMode(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>&#160; CUBLAS_POINTER_MODE_HOST);</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span>&#160; }</div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span>&#160;};</div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span>&#160;</div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span>&#160;<span class="keyword">template</span>&lt;&gt;</div><div class="line"><a name="l00735"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html"> 735</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine</a>&lt;<a class="code" href="structmshadow_1_1gpu.html">gpu</a>, double&gt; {</div><div class="line"><a name="l00736"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a39560dc0d0bed87d5f45fdb3b1b1bea3"> 736</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> cublasOperation_t <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a39560dc0d0bed87d5f45fdb3b1b1bea3">GetT</a>(<span class="keywordtype">bool</span> t) {</div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span>&#160; <span class="keywordflow">return</span> t ? CUBLAS_OP_T : CUBLAS_OP_N;</div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span>&#160; }</div><div class="line"><a name="l00739"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a99ca28256c768db9ac23224afdcecfc9"> 739</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a99ca28256c768db9ac23224afdcecfc9">SetStream</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream) {</div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span>&#160; cublasStatus_t err = cublasSetStream(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span>&#160; <a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetStream</a>(stream));</div><div class="line"><a name="l00742"></a><span class="lineno"> 742</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: set stream fail&quot;</span>;</div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span>&#160; }</div><div class="line"><a name="l00744"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ae2e8f778dfe80538d137b4ab61db6fc8"> 744</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ae2e8f778dfe80538d137b4ab61db6fc8">gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc) {</div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>&#160; cublasStatus_t err = cublasDgemm(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00751"></a><span class="lineno"> 751</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span>&#160; A, lda, B, ldb, &amp;beta, C, ldc);</div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Dgemm fail&quot;</span>;</div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span>&#160; }</div><div class="line"><a name="l00755"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ab7bda6bddf79e362d0f036c8b886e2fa"> 755</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ab7bda6bddf79e362d0f036c8b886e2fa">batched_gemm</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>&#160; <span class="keywordtype">bool</span> transa, <span class="keywordtype">bool</span> transb,</div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">int</span> k, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keyword">const</span> <span class="keywordtype">double</span> *B, <span class="keywordtype">int</span> ldb,</div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *C, <span class="keywordtype">int</span> ldc, <span class="keywordtype">int</span> batch_count,</div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span>&#160; <span class="keywordtype">double</span> **workspace) {</div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span>&#160;<span class="preprocessor">#if defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 4010 &amp;&amp; CUDA_VERSION &lt; 8000</span></div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>&#160; <span class="comment">// Cast DType* to DType** using workspace as a buffer</span></div><div class="line"><a name="l00763"></a><span class="lineno"> 763</span>&#160; <span class="keywordtype">bool</span> alloc_workspace = <span class="keyword">false</span>;</div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span>&#160; <span class="keywordflow">if</span> (workspace == NULL) {</div><div class="line"><a name="l00765"></a><span class="lineno"> 765</span>&#160; <span class="comment">// Allocate the workspace if it&#39;s NULL.</span></div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span>&#160; <span class="comment">// TODO(sxjscience) Try to move the allocation inside Tensor, which is thread-safe.</span></div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>&#160; cudaMalloc(reinterpret_cast&lt;void**&gt;(&amp;workspace), 3 * batch_count * <span class="keyword">sizeof</span>(<span class="keywordtype">double</span>*));</div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span>&#160; alloc_workspace = <span class="keyword">true</span>;</div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>&#160; }</div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace, const_cast&lt;double*&gt;(A), batch_count, m * k, stream);</div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace + batch_count,</div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span>&#160; const_cast&lt;double*&gt;(B), batch_count, k * n, stream);</div><div class="line"><a name="l00773"></a><span class="lineno"> 773</span>&#160; <a class="code" href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">GetBatchedView</a>(workspace + 2 * batch_count, C, batch_count, m * n, stream);</div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span>&#160; cublasStatus_t err = cublasDgemmBatched(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00776"></a><span class="lineno"> 776</span>&#160; (<span class="keyword">const</span> <span class="keywordtype">double</span>**)workspace, lda,</div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span>&#160; (<span class="keyword">const</span> <span class="keywordtype">double</span>**)(workspace + batch_count), ldb,</div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span>&#160; &amp;beta, workspace + 2 * batch_count, ldc, batch_count);</div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: DgemmBatched fail&quot;</span>;</div><div class="line"><a name="l00780"></a><span class="lineno"> 780</span>&#160; <span class="keywordflow">if</span> (alloc_workspace) {</div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span>&#160; cudaFree(workspace);</div><div class="line"><a name="l00782"></a><span class="lineno"> 782</span>&#160; }</div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span>&#160;<span class="preprocessor">#elif defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 8000</span></div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span>&#160; cublasStatus_t err = cublasDgemmStridedBatched(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span>&#160; GetT(transa), GetT(transb), m, n, k, &amp;alpha,</div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span>&#160; A, lda, m * k,</div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span>&#160; B, ldb, k * n,</div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>&#160; &amp;beta, C, ldc, m * n,</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span>&#160; batch_count);</div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: DgemmStridedBatched fail&quot;</span>;</div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00792"></a><span class="lineno"> 792</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span>&#160; gemm(stream, transa, transb, m, n, k, alpha,</div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span>&#160; A + i * m * k, lda, B + i * k * n, ldb,</div><div class="line"><a name="l00795"></a><span class="lineno"> 795</span>&#160; beta, C + i * m * n, ldc);</div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span>&#160; }</div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span>&#160;<span class="preprocessor">#endif // defined(__CUDACC__) &amp;&amp; CUDA_VERSION &gt;= 4010</span></div><div class="line"><a name="l00798"></a><span class="lineno"> 798</span>&#160; }</div><div class="line"><a name="l00799"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a24307fc3e49744062ebfeec854e5cdf4"> 799</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a24307fc3e49744062ebfeec854e5cdf4">gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00800"></a><span class="lineno"> 800</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00803"></a><span class="lineno"> 803</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY) {</div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>&#160; cublasStatus_t err = cublasDgemv(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>&#160; GetT(trans), m, n, &amp;alpha, A, lda, X, incX, &amp;beta, Y, incY);</div><div class="line"><a name="l00806"></a><span class="lineno"> 806</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Dgemv fail&quot;</span>;</div><div class="line"><a name="l00807"></a><span class="lineno"> 807</span>&#160; }</div><div class="line"><a name="l00808"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a7f03610034a0fb21ae664b7aca9ad242"> 808</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a7f03610034a0fb21ae664b7aca9ad242">batched_gemv</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00809"></a><span class="lineno"> 809</span>&#160; <span class="keywordtype">bool</span> trans, <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00810"></a><span class="lineno"> 810</span>&#160; <span class="keywordtype">double</span> alpha, <span class="keyword">const</span> <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda,</div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>&#160; <span class="keywordtype">double</span> beta, <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00814"></a><span class="lineno"> 814</span>&#160; gemv(stream, trans, m, n, alpha, A + i * m * n, lda,</div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span>&#160; X + i * (trans ? m : n) * incX, incX,</div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span>&#160; beta, Y + i * (trans ? n : m) * incY, incY);</div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>&#160; }</div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>&#160; }</div><div class="line"><a name="l00819"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a9c69053db1cc1e1c4741c4ad474e7e88"> 819</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a9c69053db1cc1e1c4741c4ad474e7e88">ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00820"></a><span class="lineno"> 820</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00821"></a><span class="lineno"> 821</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00822"></a><span class="lineno"> 822</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda) {</div><div class="line"><a name="l00823"></a><span class="lineno"> 823</span>&#160; cublasStatus_t err = cublasDger(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00824"></a><span class="lineno"> 824</span>&#160; m, n, &amp;alpha, X, incX, Y, incY, A, lda);</div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Dger fail&quot;</span>;</div><div class="line"><a name="l00826"></a><span class="lineno"> 826</span>&#160; }</div><div class="line"><a name="l00827"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a93988dc44c82ade1fb9e8b4b838089f7"> 827</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a93988dc44c82ade1fb9e8b4b838089f7">batched_ger</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00828"></a><span class="lineno"> 828</span>&#160; <span class="keywordtype">int</span> m, <span class="keywordtype">int</span> n, <span class="keywordtype">double</span> alpha,</div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span> *Y, <span class="keywordtype">int</span> incY, <span class="keywordtype">double</span> *A, <span class="keywordtype">int</span> lda, <span class="keywordtype">int</span> batch_count) {</div><div class="line"><a name="l00831"></a><span class="lineno"> 831</span>&#160; <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i &lt; batch_count; ++i) {</div><div class="line"><a name="l00832"></a><span class="lineno"> 832</span>&#160; ger(stream, m, n, alpha, X + i * m * incX, incX, Y + i * n * incY, incY,</div><div class="line"><a name="l00833"></a><span class="lineno"> 833</span>&#160; A + i * lda * n, lda);</div><div class="line"><a name="l00834"></a><span class="lineno"> 834</span>&#160; }</div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>&#160; }</div><div class="line"><a name="l00836"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a15c71fc2ef056a49cde2803896b2ff3c"> 836</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a15c71fc2ef056a49cde2803896b2ff3c">dot</a>(<a class="code" href="structmshadow_1_1Stream_3_01gpu_01_4.html">Stream&lt;gpu&gt;</a> *stream,</div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>&#160; <span class="keywordtype">int</span> n,</div><div class="line"><a name="l00838"></a><span class="lineno"> 838</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* X, <span class="keywordtype">int</span> incX,</div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span>&#160; <span class="keyword">const</span> <span class="keywordtype">double</span>* Y, <span class="keywordtype">int</span> incY,</div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>&#160; <span class="keywordtype">double</span> *ret) {</div><div class="line"><a name="l00841"></a><span class="lineno"> 841</span>&#160; cublasSetPointerMode(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00842"></a><span class="lineno"> 842</span>&#160; CUBLAS_POINTER_MODE_DEVICE);</div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span>&#160; cublasStatus_t err = cublasDdot(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00844"></a><span class="lineno"> 844</span>&#160; n, X, incX, Y, incY, ret);</div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>&#160; CHECK_EQ(err, CUBLAS_STATUS_SUCCESS) &lt;&lt; <span class="stringliteral">&quot;Cublas: Dot fail&quot;</span>;</div><div class="line"><a name="l00846"></a><span class="lineno"> 846</span>&#160; cublasSetPointerMode(<a class="code" href="structmshadow_1_1Stream.html">Stream&lt;gpu&gt;::GetBlasHandle</a>(stream),</div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>&#160; CUBLAS_POINTER_MODE_HOST);</div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>&#160; }</div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span>&#160;};</div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span>&#160;<span class="preprocessor">#endif // MSHADOW_USE_CUDA</span></div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>&#160;<span class="comment">// helper function to decide which shape we are in</span></div><div class="line"><a name="l00852"></a><span class="lineno"><a class="line" href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2"> 852</a></span>&#160;<span class="keyword">inline</span> <a class="code" href="structmshadow_1_1Shape.html">Shape&lt;2&gt;</a> <a class="code" href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2">GetShape</a>(<span class="keyword">const</span> <a class="code" href="structmshadow_1_1Shape.html">Shape&lt;2&gt;</a> &amp;shape, <span class="keywordtype">bool</span> <a class="code" href="namespacemshadow_1_1expr.html#afc62edfb800bb19e201b20b444831af3">transpose</a>) {</div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>&#160; <span class="keywordflow">return</span> transpose ? <a class="code" href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">Shape2</a>(shape[1], shape[0]) : shape;</div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span>&#160;}</div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>&#160;<span class="comment">// dst = dot(lhs[.T], rhs[.T])</span></div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> SV, <span class="keyword">typename</span> xpu,</div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span>&#160; <span class="keywordtype">bool</span> transpose_left, <span class="keywordtype">bool</span> transpose_right, <span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00858"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_012_00_012_00_01transpose__left_00_d9efaf4487e04a75eba0d7524dc2b6b1.html"> 858</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1DotEngine.html">DotEngine</a>&lt;SV, xpu, 2, 2, 2, transpose_left, transpose_right, DType&gt; {</div><div class="line"><a name="l00859"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_012_00_012_00_01transpose__left_00_d9efaf4487e04a75eba0d7524dc2b6b1.html#a3c3843f9868a7dbedfe22a48f821ef75"> 859</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_012_00_012_00_01transpose__left_00_d9efaf4487e04a75eba0d7524dc2b6b1.html#a3c3843f9868a7dbedfe22a48f821ef75">Eval</a>(<a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> *p_dst,</div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> &amp;lhs,</div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> &amp;rhs,</div><div class="line"><a name="l00862"></a><span class="lineno"> 862</span>&#160; DType scale) {</div><div class="line"><a name="l00863"></a><span class="lineno"> 863</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> &amp;dst = *p_dst;</div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>&#160;<span class="preprocessor">#if MSHADOW_STAND_ALONE</span></div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>&#160; <span class="keywordflow">if</span> (xpu::kDevMask == <a class="code" href="structmshadow_1_1cpu.html#a57f74b500825e21c0c246ad5b4e5016b">cpu::kDevMask</a> &amp;&amp; scale == 1.0f) {</div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>&#160; <span class="keywordflow">if</span> (!transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (!transpose_left &amp;&amp; transpose_right) {</div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs, rhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>()); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>&#160; } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (transpose_left &amp;&amp; !transpose_right) {</div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>&#160; dst = <a class="code" href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">expr::implicit_dot</a>(lhs.<a class="code" href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">T</a>(), rhs); <span class="keywordflow">return</span>;</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>&#160; }</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>&#160; }</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>&#160; <span class="comment">// set kernel stream</span></div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span>&#160; <span class="comment">// if there is no stream, crush</span></div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::SetStream</a>(dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>);</div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>&#160; <a class="code" href="structmshadow_1_1Shape.html">Shape&lt;2&gt;</a> sleft = <a class="code" href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2">GetShape</a>(lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a>, transpose_left);</div><div class="line"><a name="l00879"></a><span class="lineno"> 879</span>&#160; <a class="code" href="structmshadow_1_1Shape.html">Shape&lt;2&gt;</a> sright = <a class="code" href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2">GetShape</a>(rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a>, transpose_right);</div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span>&#160; CHECK(dst.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) == sleft[0] &amp;&amp; dst.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1) == sright[1] &amp;&amp; sleft[1] == sright[0])</div><div class="line"><a name="l00881"></a><span class="lineno"> 881</span>&#160; &lt;&lt; <span class="stringliteral">&quot;dot-gemm: matrix shape mismatch&quot;</span>;</div><div class="line"><a name="l00882"></a><span class="lineno"> 882</span>&#160; <span class="comment">// use column major argument to compatible with most BLAS</span></div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::gemm</a></div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span>&#160; (dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>,</div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span>&#160; transpose_right , transpose_left,</div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span>&#160; transpose_right ? rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) : rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1),</div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span>&#160; transpose_left ? lhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1) : lhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0),</div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span>&#160; transpose_right ? rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1) : rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0),</div><div class="line"><a name="l00889"></a><span class="lineno"> 889</span>&#160; DType(scale * SV::AlphaBLAS()),</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</span>&#160; rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, rhs.<a class="code" href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">stride_</a>,</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>&#160; lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, lhs.<a class="code" href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">stride_</a>,</div><div class="line"><a name="l00892"></a><span class="lineno"> 892</span>&#160; DType(SV::BetaBLAS()),</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span>&#160; dst.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, dst.<a class="code" href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">stride_</a>);</div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>&#160; }</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>&#160;};</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> SV, <span class="keyword">typename</span> xpu, <span class="keywordtype">bool</span> transpose_right, <span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00897"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_011_00_011_00_012_00_01false_00_01transpose__right_00_01DType_01_4.html"> 897</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1DotEngine.html">DotEngine</a>&lt;SV, xpu, 1, 1, 2, false, transpose_right, DType&gt; {</div><div class="line"><a name="l00898"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_011_00_011_00_012_00_01false_00_01transpose__right_00_01DType_01_4.html#aee303d45ee3daf57d08fd1ada06571fc"> 898</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_011_00_011_00_012_00_01false_00_01transpose__right_00_01DType_01_4.html#aee303d45ee3daf57d08fd1ada06571fc">Eval</a>(<a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 1, DType&gt;</a> *p_dst,</div><div class="line"><a name="l00899"></a><span class="lineno"> 899</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 1, DType&gt;</a> &amp;lhs,</div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> &amp;rhs,</div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>&#160; DType scale) {</div><div class="line"><a name="l00902"></a><span class="lineno"> 902</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 1, DType&gt;</a> &amp;dst = *p_dst;</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>&#160; <span class="comment">// set kernel stream</span></div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span>&#160; <span class="comment">// if there is no stream, crush</span></div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::SetStream</a>(dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>);</div><div class="line"><a name="l00906"></a><span class="lineno"> 906</span>&#160; <a class="code" href="structmshadow_1_1Shape.html">Shape&lt;2&gt;</a> sright = <a class="code" href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2">GetShape</a>(rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a>, transpose_right);</div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>&#160; CHECK(dst.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) == sright[1] &amp;&amp; lhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) == sright[0])</div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>&#160; &lt;&lt; <span class="stringliteral">&quot;dot-gemv: matrix shape mismatch&quot;</span></div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span>&#160; &lt;&lt; <span class="stringliteral">&quot;dst: &quot;</span> &lt;&lt; dst.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a> &lt;&lt; <span class="stringliteral">&quot;\n&quot;</span></div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span>&#160; &lt;&lt; <span class="stringliteral">&quot;lhs: &quot;</span> &lt;&lt; lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a> &lt;&lt; <span class="stringliteral">&quot;\n&quot;</span></div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span>&#160; &lt;&lt; <span class="stringliteral">&quot;rhs: &quot;</span> &lt;&lt; sright &lt;&lt; <span class="stringliteral">&quot;\n&quot;</span>;</div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::gemv</a></div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span>&#160; (dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>,</div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span>&#160; transpose_right,</div><div class="line"><a name="l00915"></a><span class="lineno"> 915</span>&#160; rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1), rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0), scale * SV::AlphaBLAS(),</div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span>&#160; rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, rhs.<a class="code" href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">stride_</a>,</div><div class="line"><a name="l00917"></a><span class="lineno"> 917</span>&#160; lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, 1, SV::BetaBLAS(),</div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>&#160; dst.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, 1);</div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>&#160; }</div><div class="line"><a name="l00920"></a><span class="lineno"> 920</span>&#160;};</div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span>&#160;<span class="keyword">template</span>&lt;<span class="keyword">typename</span> SV, <span class="keyword">typename</span> xpu, <span class="keyword">typename</span> DType&gt;</div><div class="line"><a name="l00922"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_011_00_011_00_01true_00_01false_00_01DType_01_4.html"> 922</a></span>&#160;<span class="keyword">struct </span><a class="code" href="structmshadow_1_1expr_1_1DotEngine.html">DotEngine</a>&lt;SV, xpu, 2, 1, 1, true, false, DType&gt; {</div><div class="line"><a name="l00923"></a><span class="lineno"><a class="line" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_011_00_011_00_01true_00_01false_00_01DType_01_4.html#a5dd265709fa1d32caa8c351741364fc5"> 923</a></span>&#160; <span class="keyword">inline</span> <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_011_00_011_00_01true_00_01false_00_01DType_01_4.html#a5dd265709fa1d32caa8c351741364fc5">Eval</a>(<a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> *p_dst,</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 1, DType&gt;</a> &amp;lhs,</div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>&#160; <span class="keyword">const</span> <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 1, DType&gt;</a> &amp;rhs,</div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>&#160; DType scale) {</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span>&#160; <a class="code" href="structmshadow_1_1Tensor.html">Tensor&lt;xpu, 2, DType&gt;</a> &amp;dst = *p_dst;</div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span>&#160; <span class="comment">// set kernel stream</span></div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>&#160; <span class="comment">// if there is no stream, crush</span></div><div class="line"><a name="l00930"></a><span class="lineno"> 930</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::SetStream</a>(dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>);</div><div class="line"><a name="l00931"></a><span class="lineno"> 931</span>&#160; CHECK(dst.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) == lhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0) &amp;&amp; dst.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(1) == rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0))</div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>&#160; &lt;&lt; <span class="stringliteral">&quot;dot-ger: matrix shape mismatch&quot;</span></div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>&#160; &lt;&lt; <span class="stringliteral">&quot;dst: &quot;</span> &lt;&lt; dst.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a> &lt;&lt; <span class="stringliteral">&quot;\n&quot;</span></div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>&#160; &lt;&lt; <span class="stringliteral">&quot;lhs: &quot;</span> &lt;&lt; lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a> &lt;&lt; <span class="stringliteral">&quot;\n&quot;</span></div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>&#160; &lt;&lt; <span class="stringliteral">&quot;rhs: &quot;</span> &lt;&lt; rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">shape_</a>;</div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>&#160; if (SV::BetaBLAS() == 0.0f) {</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1BLASEngine.html">BLASEngine&lt;xpu, DType&gt;::ger</a></div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span>&#160; (dst.<a class="code" href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">stream_</a>, rhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0), lhs.<a class="code" href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">size</a>(0), scale * SV::AlphaBLAS(),</div><div class="line"><a name="l00939"></a><span class="lineno"> 939</span>&#160; rhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, 1, lhs.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, 1, dst.<a class="code" href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">dptr_</a>, dst.<a class="code" href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">stride_</a>);</div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span>&#160; } <span class="keywordflow">else</span> {</div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>&#160; <a class="code" href="structmshadow_1_1expr_1_1DotEngine.html">DotEngine</a>&lt;SV, xpu, 2, 2, 2, <span class="keyword">true</span>, <span class="keyword">false</span>,</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>&#160; DType&gt;::Eval(p_dst, lhs.<a class="code" href="structmshadow_1_1Tensor.html#a48a5927e810fbc45e43e92cfe397d9f2">FlatTo2D</a>(), rhs.<a class="code" href="structmshadow_1_1Tensor.html#a48a5927e810fbc45e43e92cfe397d9f2">FlatTo2D</a>(), scale);</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>&#160; }</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>&#160; }</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>&#160;};</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>&#160;} <span class="comment">// namespace expr</span></div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>&#160;} <span class="comment">// namespace mshadow</span></div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span>&#160;<span class="preprocessor">#endif // MSHADOW_DOT_ENGINE_INL_H_</span></div><div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a6d1f6461bd234ab76f380ac6da8f8f1c"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a6d1f6461bd234ab76f380ac6da8f8f1c">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::ger</a></div><div class="ttdeci">static void ger(Stream&lt; gpu &gt; *stream, int m, int n, half::half_t alpha, const half::half_t *X, int incX, const half::half_t *Y, int incY, half::half_t *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:596</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_aea1017183bb11680f4f041c13c5d3a7f"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#aea1017183bb11680f4f041c13c5d3a7f">mshadow::expr::BLASEngine::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; Device &gt; *stream, bool trans, int m, int n, DType alpha, const DType *A, int lda, const DType *X, int incX, DType beta, DType *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:106</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_a85dfe7b9fe210cb4dbe06a97fa81ecda"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#a85dfe7b9fe210cb4dbe06a97fa81ecda">mshadow::expr::BLASEngine::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; Device &gt; *stream, bool transa, bool transb, int m, int n, int k, DType alpha, const DType *A, int lda, const DType *B, int ldb, DType beta, DType *C, int ldc, int batch_count, DType **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:91</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a99ca28256c768db9ac23224afdcecfc9"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a99ca28256c768db9ac23224afdcecfc9">mshadow::expr::BLASEngine&lt; gpu, double &gt;::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; gpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:739</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_ad88269ee104f5310e0a5ccc6b1fb1339"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ad88269ee104f5310e0a5ccc6b1fb1339">mshadow::expr::BLASEngine&lt; cpu, double &gt;::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; cpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:406</div></div>
<div class="ttc" id="namespacemshadow_1_1expr_html_aef7b5fe2e264d0bb3a83de21836acd99"><div class="ttname"><a href="namespacemshadow_1_1expr.html#aef7b5fe2e264d0bb3a83de21836acd99">mshadow::expr::implicit_dot</a></div><div class="ttdeci">ImplicitGEMMExp&lt; LhsExp, RhsExp, DType &gt; implicit_dot(const Exp&lt; LhsExp, DType, e1 &gt; &amp;lhs, const Exp&lt; RhsExp, DType, e2 &gt; &amp;rhs)</div><div class="ttdef"><b>Definition:</b> implicit_gemm.h:64</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_abdf3b3f658b9d8277c1351229ad3e6ce"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#abdf3b3f658b9d8277c1351229ad3e6ce">mshadow::expr::BLASEngine&lt; cpu, float &gt;::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; cpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:297</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_ab384dfeab544e586a57258b0ce3082c2"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ab384dfeab544e586a57258b0ce3082c2">mshadow::expr::BLASEngine&lt; gpu, float &gt;::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; gpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:622</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a979610afce873cc7d4bac17a22a7518a"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a979610afce873cc7d4bac17a22a7518a">mshadow::expr::BLASEngine&lt; gpu, float &gt;::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, float alpha, const float *A, int lda, const float *X, int incX, float beta, float *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:691</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_ad86d6759c585efb5229b3a0659973838"><div class="ttname"><a href="structmshadow_1_1Tensor.html#ad86d6759c585efb5229b3a0659973838">mshadow::Tensor::dptr_</a></div><div class="ttdeci">DType * dptr_</div><div class="ttdoc">pointer to the data </div><div class="ttdef"><b>Definition:</b> tensor.h:434</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_ae014aad49673f34d8f4149ea4525820b"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ae014aad49673f34d8f4149ea4525820b">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, half::half_t alpha, const half::half_t *A, int lda, const half::half_t *X, int incX, half::half_t beta, half::half_t *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:582</div></div>
<div class="ttc" id="namespacemshadow_1_1expr_html_a1552427378286a426185fe2c503c8ad2"><div class="ttname"><a href="namespacemshadow_1_1expr.html#a1552427378286a426185fe2c503c8ad2">mshadow::expr::GetShape</a></div><div class="ttdeci">Shape&lt; 2 &gt; GetShape(const Shape&lt; 2 &gt; &amp;shape, bool transpose)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:852</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_a19e843390ec50c3064698408603fc63a"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a19e843390ec50c3064698408603fc63a">mshadow::expr::BLASEngine&lt; cpu, double &gt;::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; cpu &gt; *stream, bool transa, bool transb, int m, int n, int k, double alpha, const double *A, int lda, const double *B, int ldb, double beta, double *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:408</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a5c9f30798d4e51a1f989621063e2e0ad"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5c9f30798d4e51a1f989621063e2e0ad">mshadow::expr::BLASEngine&lt; cpu, float &gt;::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; cpu &gt; *stream, bool trans, int m, int n, float alpha, const float *A, int lda, const float *X, int incX, float beta, float *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:358</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a24307fc3e49744062ebfeec854e5cdf4"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a24307fc3e49744062ebfeec854e5cdf4">mshadow::expr::BLASEngine&lt; gpu, double &gt;::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, double alpha, const double *A, int lda, const double *X, int incX, double beta, double *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:799</div></div>
<div class="ttc" id="structmshadow_1_1Shape_html"><div class="ttname"><a href="structmshadow_1_1Shape.html">mshadow::Shape&lt; 2 &gt;</a></div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_a8a5491682d961329ceed384c6f8c9931"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a8a5491682d961329ceed384c6f8c9931">mshadow::expr::BLASEngine&lt; cpu, double &gt;::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; cpu &gt; *stream, int m, int n, double alpha, const double *X, int incX, const double *Y, int incY, double *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:492</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_ab7bda6bddf79e362d0f036c8b886e2fa"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ab7bda6bddf79e362d0f036c8b886e2fa">mshadow::expr::BLASEngine&lt; gpu, double &gt;::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, double alpha, const double *A, int lda, const double *B, int ldb, double beta, double *C, int ldc, int batch_count, double **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:755</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_aefac67f858a67b1bba74bc06ed2c3e99"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aefac67f858a67b1bba74bc06ed2c3e99">mshadow::expr::BLASEngine&lt; cpu, double &gt;::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; cpu &gt; *stream, bool trans, int m, int n, double alpha, const double *A, int lda, const double *X, int incX, double beta, double *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:467</div></div>
<div class="ttc" id="structmshadow_1_1Stream_3_01gpu_01_4_html"><div class="ttname"><a href="structmshadow_1_1Stream_3_01gpu_01_4.html">mshadow::Stream&lt; gpu &gt;</a></div><div class="ttdef"><b>Definition:</b> stream_gpu-inl.h:37</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a7f03610034a0fb21ae664b7aca9ad242"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a7f03610034a0fb21ae664b7aca9ad242">mshadow::expr::BLASEngine&lt; gpu, double &gt;::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, double alpha, const double *A, int lda, const double *X, int incX, double beta, double *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:808</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a8895cfca8053511f48053a093d43a7f6"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a8895cfca8053511f48053a093d43a7f6">mshadow::expr::BLASEngine&lt; cpu, float &gt;::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; cpu &gt; *stream, int m, int n, float alpha, const float *X, int incX, const float *Y, int incY, float *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:383</div></div>
<div class="ttc" id="implicit__gemm_8h_html"><div class="ttname"><a href="implicit__gemm_8h.html">implicit_gemm.h</a></div><div class="ttdoc">support for implicit GEMM operation </div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_af75b98880cf90282c31c002966e2c3ca"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#af75b98880cf90282c31c002966e2c3ca">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; gpu &gt; *stream, int m, int n, half::half_t alpha, const half::half_t *X, int incX, const half::half_t *Y, int incY, half::half_t *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:602</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_ad10c7414c5948e789e8761df2083c4e5"><div class="ttname"><a href="structmshadow_1_1Tensor.html#ad10c7414c5948e789e8761df2083c4e5">mshadow::Tensor::shape_</a></div><div class="ttdeci">Shape&lt; dimension &gt; shape_</div><div class="ttdoc">shape of the tensor </div><div class="ttdef"><b>Definition:</b> tensor.h:436</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_ad3f8beddfbc19c0a25be5f856a540f39"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ad3f8beddfbc19c0a25be5f856a540f39">mshadow::expr::BLASEngine&lt; cpu, float &gt;::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; cpu &gt; *stream, bool transa, bool transb, int m, int n, int k, float alpha, const float *A, int lda, const float *B, int ldb, float beta, float *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:299</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a1ef82a21b3debea986511176e88163f5"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a1ef82a21b3debea986511176e88163f5">mshadow::expr::BLASEngine&lt; gpu, float &gt;::GetT</a></div><div class="ttdeci">static cublasOperation_t GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:619</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a7dbffea32aea2ed9b58e8098c80cd05f"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a7dbffea32aea2ed9b58e8098c80cd05f">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, half::half_t alpha, const half::half_t *A, int lda, const half::half_t *X, int incX, half::half_t beta, half::half_t *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:589</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a3b5d0cfe43d79f1b8aadaca3ae0cf0cc"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a3b5d0cfe43d79f1b8aadaca3ae0cf0cc">mshadow::expr::BLASEngine&lt; gpu, float &gt;::ger</a></div><div class="ttdeci">static void ger(Stream&lt; gpu &gt; *stream, int m, int n, float alpha, const float *X, int incX, const float *Y, int incY, float *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:702</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a93988dc44c82ade1fb9e8b4b838089f7"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a93988dc44c82ade1fb9e8b4b838089f7">mshadow::expr::BLASEngine&lt; gpu, double &gt;::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; gpu &gt; *stream, int m, int n, double alpha, const double *X, int incX, const double *Y, int incY, double *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:827</div></div>
<div class="ttc" id="namespacemshadow_1_1expr_html_a496d2f5fb874d0a8d59b6a0b4374e2df"><div class="ttname"><a href="namespacemshadow_1_1expr.html#a496d2f5fb874d0a8d59b6a0b4374e2df">mshadow::expr::dot</a></div><div class="ttdeci">DotExp&lt; TA, TB, false, false, DType &gt; dot(const RValueExp&lt; TA, DType &gt; &amp;lhs, const RValueExp&lt; TB, DType &gt; &amp;rhs)</div><div class="ttdoc">dot operator def </div><div class="ttdef"><b>Definition:</b> expression.h:240</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_aa8c29bbc3c224b7da63a09d1f3d7727e"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa8c29bbc3c224b7da63a09d1f3d7727e">mshadow::expr::BLASEngine&lt; cpu, double &gt;::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; cpu &gt; *stream, bool trans, int m, int n, double alpha, const double *A, int lda, const double *X, int incX, double beta, double *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:475</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a463ff27ac0c305235aee476f5d6b7608"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a463ff27ac0c305235aee476f5d6b7608">mshadow::expr::BLASEngine&lt; gpu, float &gt;::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, float alpha, const float *A, int lda, const float *B, int ldb, float beta, float *C, int ldc, int batch_count, float **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:638</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1DotEngine_html"><div class="ttname"><a href="structmshadow_1_1expr_1_1DotEngine.html">mshadow::expr::DotEngine</a></div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:70</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a76cdb254a4f93f1e1fbaaa335177bccc"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a76cdb254a4f93f1e1fbaaa335177bccc">mshadow::expr::BLASEngine&lt; gpu, float &gt;::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; gpu &gt; *stream, int m, int n, float alpha, const float *X, int incX, const float *Y, int incY, float *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:710</div></div>
<div class="ttc" id="structmshadow_1_1Stream_3_01gpu_01_4_html_a4ec551c8440da0c89eb728e753906936"><div class="ttname"><a href="structmshadow_1_1Stream_3_01gpu_01_4.html#a4ec551c8440da0c89eb728e753906936">mshadow::Stream&lt; gpu &gt;::prop</a></div><div class="ttdeci">cudaDeviceProp prop</div><div class="ttdoc">cudaDeviceProp </div><div class="ttdef"><b>Definition:</b> stream_gpu-inl.h:62</div></div>
<div class="ttc" id="structmshadow_1_1gpu_html"><div class="ttname"><a href="structmshadow_1_1gpu.html">mshadow::gpu</a></div><div class="ttdoc">device name GPU </div><div class="ttdef"><b>Definition:</b> tensor.h:46</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_abd8987afc8ece12882f139f39c8fca7a"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#abd8987afc8ece12882f139f39c8fca7a">mshadow::expr::BLASEngine::batched_ger</a></div><div class="ttdeci">static void batched_ger(Stream&lt; Device &gt; *stream, int m, int n, DType alpha, const DType *X, int incX, const DType *Y, int incY, DType *A, int lda, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:119</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a39560dc0d0bed87d5f45fdb3b1b1bea3"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a39560dc0d0bed87d5f45fdb3b1b1bea3">mshadow::expr::BLASEngine&lt; gpu, double &gt;::GetT</a></div><div class="ttdeci">static cublasOperation_t GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:736</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_012_00_012_00_01transpose__left_00_d9efaf4487e04a75eba0d7524dc2b6b1_html_a3c3843f9868a7dbedfe22a48f821ef75"><div class="ttname"><a href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_012_00_012_00_01transpose__left_00_d9efaf4487e04a75eba0d7524dc2b6b1.html#a3c3843f9868a7dbedfe22a48f821ef75">mshadow::expr::DotEngine&lt; SV, xpu, 2, 2, 2, transpose_left, transpose_right, DType &gt;::Eval</a></div><div class="ttdeci">static void Eval(Tensor&lt; xpu, 2, DType &gt; *p_dst, const Tensor&lt; xpu, 2, DType &gt; &amp;lhs, const Tensor&lt; xpu, 2, DType &gt; &amp;rhs, DType scale)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:859</div></div>
<div class="ttc" id="classmshadow_1_1expr_1_1RValueExp_html_afbddc739cc4f38e6509be4fcab0c6158"><div class="ttname"><a href="classmshadow_1_1expr_1_1RValueExp.html#afbddc739cc4f38e6509be4fcab0c6158">mshadow::expr::RValueExp&lt; Tensor&lt; Device, dimension, DType &gt;, DType &gt;::T</a></div><div class="ttdeci">const TransposeExp&lt; Tensor&lt; Device, dimension, DType &gt;, DType &gt; T(void) const</div><div class="ttdoc">transpose of a matrix </div><div class="ttdef"><b>Definition:</b> expression.h:154</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_a48a5927e810fbc45e43e92cfe397d9f2"><div class="ttname"><a href="structmshadow_1_1Tensor.html#a48a5927e810fbc45e43e92cfe397d9f2">mshadow::Tensor::FlatTo2D</a></div><div class="ttdeci">MSHADOW_XINLINE Tensor&lt; Device, 2, DType &gt; FlatTo2D(void) const</div><div class="ttdoc">flatten the tensor to 2 dimension, collapse the higher dimensions together </div><div class="ttdef"><b>Definition:</b> tensor.h:519</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a8d6d8e682966822bb85b6ff6fee32ce2"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a8d6d8e682966822bb85b6ff6fee32ce2">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, half::half_t alpha, const half::half_t *A, int lda, const half::half_t *B, int ldb, half::half_t beta, half::half_t *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:523</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_a88cbcae11653307bfa4c99804320b638"><div class="ttname"><a href="structmshadow_1_1Tensor.html#a88cbcae11653307bfa4c99804320b638">mshadow::Tensor::size</a></div><div class="ttdeci">MSHADOW_XINLINE index_t size(int idx) const</div><div class="ttdoc">return size of i-th dimension, start counting from highest dimension </div><div class="ttdef"><b>Definition:</b> tensor.h:505</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a517affe439b1378dff42a2efe8c21dd8"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a517affe439b1378dff42a2efe8c21dd8">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, half::half_t alpha, const half::half_t *A, int lda, const half::half_t *B, int ldb, half::half_t beta, half::half_t *C, int ldc, int batch_count, half::half_t **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:550</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_a621f673c37d15f9e1acb76bd7be3be82"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a621f673c37d15f9e1acb76bd7be3be82">mshadow::expr::BLASEngine&lt; cpu, double &gt;::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; cpu &gt; *stream, bool transa, bool transb, int m, int n, int k, double alpha, const double *A, int lda, const double *B, int ldb, double beta, double *C, int ldc, int batch_count, double **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:416</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_ae2e8f778dfe80538d137b4ab61db6fc8"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#ae2e8f778dfe80538d137b4ab61db6fc8">mshadow::expr::BLASEngine&lt; gpu, double &gt;::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, double alpha, const double *A, int lda, const double *B, int ldb, double beta, double *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:744</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_a470e53711519b24cad68bbd41d600052"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#a470e53711519b24cad68bbd41d600052">mshadow::expr::BLASEngine&lt; cpu, double &gt;::GetT</a></div><div class="ttdeci">static CBLAS_TRANSPOSE GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:403</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_af21b0de1c3d687671e36cffb464d5988"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#af21b0de1c3d687671e36cffb464d5988">mshadow::expr::BLASEngine::dot</a></div><div class="ttdeci">static void dot(Stream&lt; Device &gt; *stream, int n, const DType *X, int incX, const DType *Y, int incY, DType *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:125</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a9977ea76d73267cf26680ec0b34ab062"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a9977ea76d73267cf26680ec0b34ab062">mshadow::expr::BLASEngine&lt; cpu, float &gt;::batched_gemm</a></div><div class="ttdeci">static void batched_gemm(Stream&lt; cpu &gt; *stream, bool transa, bool transb, int m, int n, int k, float alpha, const float *A, int lda, const float *B, int ldb, float beta, float *C, int ldc, int batch_count, float **workspace)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:307</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_ac742f67973dd9ce1c444f0c476749d10"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#ac742f67973dd9ce1c444f0c476749d10">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::GetT</a></div><div class="ttdeci">static cublasOperation_t GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:515</div></div>
<div class="ttc" id="namespacemshadow_html_ad2853583180a7dbe9cbf6b0e15ae60ef"><div class="ttname"><a href="namespacemshadow.html#ad2853583180a7dbe9cbf6b0e15ae60ef">mshadow::Shape2</a></div><div class="ttdeci">MSHADOW_XINLINE Shape&lt; 2 &gt; Shape2(index_t s0, index_t s1)</div><div class="ttdoc">construct a two dimension shape, stride will equal s0 </div><div class="ttdef"><b>Definition:</b> tensor.h:216</div></div>
<div class="ttc" id="structmshadow_1_1cpu_html_a57f74b500825e21c0c246ad5b4e5016b"><div class="ttname"><a href="structmshadow_1_1cpu.html#a57f74b500825e21c0c246ad5b4e5016b">mshadow::cpu::kDevMask</a></div><div class="ttdeci">static const int kDevMask</div><div class="ttdoc">device flag number, identifies this device </div><div class="ttdef"><b>Definition:</b> tensor.h:43</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_ab923458f56a77bc3e68d541f85ce8599"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#ab923458f56a77bc3e68d541f85ce8599">mshadow::expr::BLASEngine&lt; cpu, double &gt;::ger</a></div><div class="ttdeci">static void ger(Stream&lt; cpu &gt; *stream, int m, int n, double alpha, const double *X, int incX, const double *Y, int incY, double *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:486</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a3d822069b1c05de013ff06e4680c6454"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a3d822069b1c05de013ff06e4680c6454">mshadow::expr::BLASEngine&lt; cpu, float &gt;::ger</a></div><div class="ttdeci">static void ger(Stream&lt; cpu &gt; *stream, int m, int n, float alpha, const float *X, int incX, const float *Y, int incY, float *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:377</div></div>
<div class="ttc" id="namespacemshadow_html_a92e5ec7d9574c70e69b97b881a688e88"><div class="ttname"><a href="namespacemshadow.html#a92e5ec7d9574c70e69b97b881a688e88">mshadow::GetBatchedView</a></div><div class="ttdeci">void GetBatchedView(DType **dst, DType *src, int num, int stride, Stream&lt; Device &gt; *stream)</div><div class="ttdoc">CPU/GPU: Get a batched view of the src array. dst[i] = src + i * stride. </div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_ae1db170ffdc4480318bbaf17fe227b4b"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#ae1db170ffdc4480318bbaf17fe227b4b">mshadow::expr::BLASEngine&lt; gpu, float &gt;::dot</a></div><div class="ttdeci">static void dot(Stream&lt; gpu &gt; *stream, int n, const float *X, int incX, const float *Y, int incY, float *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:719</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_011_00_011_00_01true_00_01false_00_01DType_01_4_html_a5dd265709fa1d32caa8c351741364fc5"><div class="ttname"><a href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_012_00_011_00_011_00_01true_00_01false_00_01DType_01_4.html#a5dd265709fa1d32caa8c351741364fc5">mshadow::expr::DotEngine&lt; SV, xpu, 2, 1, 1, true, false, DType &gt;::Eval</a></div><div class="ttdeci">static void Eval(Tensor&lt; xpu, 2, DType &gt; *p_dst, const Tensor&lt; xpu, 1, DType &gt; &amp;lhs, const Tensor&lt; xpu, 1, DType &gt; &amp;rhs, DType scale)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:923</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a5bf25e6bdccbe70d1f986701ce6414a2"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a5bf25e6bdccbe70d1f986701ce6414a2">mshadow::expr::BLASEngine&lt; cpu, float &gt;::dot</a></div><div class="ttdeci">static void dot(Stream&lt; cpu &gt; *stream, int n, const float *X, int incX, const float *Y, int incY, float *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:392</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_a3785a79812f5e01713845480a790601c"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#a3785a79812f5e01713845480a790601c">mshadow::expr::BLASEngine::ger</a></div><div class="ttdeci">static void ger(Stream&lt; Device &gt; *stream, int m, int n, DType alpha, const DType *X, int incX, const DType *Y, int incY, DType *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:113</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a15c71fc2ef056a49cde2803896b2ff3c"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a15c71fc2ef056a49cde2803896b2ff3c">mshadow::expr::BLASEngine&lt; gpu, double &gt;::dot</a></div><div class="ttdeci">static void dot(Stream&lt; gpu &gt; *stream, int n, const double *X, int incX, const double *Y, int incY, double *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:836</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_ac38c4c955ccbf0749acf51afcb5143d4"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#ac38c4c955ccbf0749acf51afcb5143d4">mshadow::expr::BLASEngine&lt; cpu, float &gt;::batched_gemv</a></div><div class="ttdeci">static void batched_gemv(Stream&lt; cpu &gt; *stream, bool trans, int m, int n, float alpha, const float *A, int lda, const float *X, int incX, float beta, float *Y, int incY, int batch_count)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:366</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_011_00_011_00_012_00_01false_00_01transpose__right_00_01DType_01_4_html_aee303d45ee3daf57d08fd1ada06571fc"><div class="ttname"><a href="structmshadow_1_1expr_1_1DotEngine_3_01SV_00_01xpu_00_011_00_011_00_012_00_01false_00_01transpose__right_00_01DType_01_4.html#aee303d45ee3daf57d08fd1ada06571fc">mshadow::expr::DotEngine&lt; SV, xpu, 1, 1, 2, false, transpose_right, DType &gt;::Eval</a></div><div class="ttdeci">static void Eval(Tensor&lt; xpu, 1, DType &gt; *p_dst, const Tensor&lt; xpu, 1, DType &gt; &amp;lhs, const Tensor&lt; xpu, 2, DType &gt; &amp;rhs, DType scale)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:898</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4_html_aa6492ff9a5684ad19cf90b875e00b365"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01double_01_4.html#aa6492ff9a5684ad19cf90b875e00b365">mshadow::expr::BLASEngine&lt; cpu, double &gt;::dot</a></div><div class="ttdeci">static void dot(Stream&lt; cpu &gt; *stream, int n, const double *X, int incX, const double *Y, int incY, double *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:501</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4_html_a9c69053db1cc1e1c4741c4ad474e7e88"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01double_01_4.html#a9c69053db1cc1e1c4741c4ad474e7e88">mshadow::expr::BLASEngine&lt; gpu, double &gt;::ger</a></div><div class="ttdeci">static void ger(Stream&lt; gpu &gt; *stream, int m, int n, double alpha, const double *X, int incX, const double *Y, int incY, double *A, int lda)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:819</div></div>
<div class="ttc" id="namespacemshadow_html"><div class="ttname"><a href="namespacemshadow.html">mshadow</a></div><div class="ttdoc">overloaded + operator between half_t and bf16_t </div><div class="ttdef"><b>Definition:</b> base.h:334</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a7b194dedaa7a20eb61fce9d6ccff7690"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a7b194dedaa7a20eb61fce9d6ccff7690">mshadow::expr::BLASEngine&lt; gpu, float &gt;::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; gpu &gt; *stream, bool trans, int m, int n, float alpha, const float *A, int lda, const float *X, int incX, float beta, float *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:682</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_a66dd260062b01a60d54370cb71871830"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#a66dd260062b01a60d54370cb71871830">mshadow::expr::BLASEngine::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; Device &gt; *stream, bool transa, bool transb, int m, int n, int k, DType alpha, const DType *A, int lda, const DType *B, int ldb, DType beta, DType *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:84</div></div>
<div class="ttc" id="namespacemshadow_1_1expr_html_afc62edfb800bb19e201b20b444831af3"><div class="ttname"><a href="namespacemshadow_1_1expr.html#afc62edfb800bb19e201b20b444831af3">mshadow::expr::transpose</a></div><div class="ttdeci">TransposeExExp&lt; SrcExp, DType, ExpInfo&lt; SrcExp &gt;::kDim &gt; transpose(const Exp&lt; SrcExp, DType, etype &gt; &amp;src, Shape&lt; ExpInfo&lt; SrcExp &gt;::kDim &gt; axes)</div><div class="ttdoc">a expression that reshapes a tensor to another shape </div><div class="ttdef"><b>Definition:</b> transpose.h:76</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_afee556f188e29bbd0ecc45fe98d3c1c3"><div class="ttname"><a href="structmshadow_1_1Tensor.html#afee556f188e29bbd0ecc45fe98d3c1c3">mshadow::Tensor::stride_</a></div><div class="ttdeci">index_t stride_</div><div class="ttdoc">storing the stride information in x dimension this is used to deal with pitch allocation in gpu or ss...</div><div class="ttdef"><b>Definition:</b> tensor.h:441</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a2091ee54add7a9ff95c6e1f08bed21d5"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a2091ee54add7a9ff95c6e1f08bed21d5">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; gpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:518</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html">mshadow::expr::BLASEngine</a></div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:78</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4_html_a88faac9a1b47990aed30b1b284154624"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01half_1_1half__t_01_4.html#a88faac9a1b47990aed30b1b284154624">mshadow::expr::BLASEngine&lt; gpu, half::half_t &gt;::dot</a></div><div class="ttdeci">static void dot(Stream&lt; gpu &gt; *stream, int n, const half::half_t *X, int incX, const half::half_t *Y, int incY, half::half_t *ret)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:608</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html"><div class="ttname"><a href="structmshadow_1_1Tensor.html">mshadow::Tensor</a></div><div class="ttdoc">general tensor </div><div class="ttdef"><b>Definition:</b> tensor.h:420</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_a6186eb5d53e0a9d956675550b5cdb41a"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#a6186eb5d53e0a9d956675550b5cdb41a">mshadow::expr::BLASEngine::gemv</a></div><div class="ttdeci">static void gemv(Stream&lt; Device &gt; *stream, bool trans, int m, int n, DType alpha, const DType *A, int lda, const DType *X, int incX, DType beta, DType *Y, int incY)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:99</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4_html_a2beab1ca667698e87fe09b6c63e22b8f"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01gpu_00_01float_01_4.html#a2beab1ca667698e87fe09b6c63e22b8f">mshadow::expr::BLASEngine&lt; gpu, float &gt;::gemm</a></div><div class="ttdeci">static void gemm(Stream&lt; gpu &gt; *stream, bool transa, bool transb, int m, int n, int k, float alpha, const float *A, int lda, const float *B, int ldb, float beta, float *C, int ldc)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:627</div></div>
<div class="ttc" id="3rdparty_2mshadow_2mshadow_2base_8h_html"><div class="ttname"><a href="3rdparty_2mshadow_2mshadow_2base_8h.html">base.h</a></div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_ac0b67fda3ecdf2087b816602ad43eda3"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#ac0b67fda3ecdf2087b816602ad43eda3">mshadow::expr::BLASEngine::SetStream</a></div><div class="ttdeci">static void SetStream(Stream&lt; Device &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:82</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_html_ac1ee5340c4bf6e123db07dfc57a952d7"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine.html#ac1ee5340c4bf6e123db07dfc57a952d7">mshadow::expr::BLASEngine::GetT</a></div><div class="ttdeci">static bool GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:79</div></div>
<div class="ttc" id="structmshadow_1_1Tensor_html_a485d1558eaa9eded0b8fd23ed82c6298"><div class="ttname"><a href="structmshadow_1_1Tensor.html#a485d1558eaa9eded0b8fd23ed82c6298">mshadow::Tensor::stream_</a></div><div class="ttdeci">Stream&lt; Device &gt; * stream_</div><div class="ttdoc">stream where the computation lies stream is a device dependency concept where each computation ...</div><div class="ttdef"><b>Definition:</b> tensor.h:446</div></div>
<div class="ttc" id="namespacemshadow_html_a9aefcfd5ca44f5e146d2274d64acc239"><div class="ttname"><a href="namespacemshadow.html#a9aefcfd5ca44f5e146d2274d64acc239">mshadow::GetBatchedView</a></div><div class="ttdeci">void GetBatchedView(DType **dst, DType *src, int num, int stride, Stream&lt; cpu &gt; *stream)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:49</div></div>
<div class="ttc" id="structmshadow_1_1Stream_html"><div class="ttname"><a href="structmshadow_1_1Stream.html">mshadow::Stream</a></div><div class="ttdoc">computaion stream structure, used for asynchronous computations </div><div class="ttdef"><b>Definition:</b> tensor.h:383</div></div>
<div class="ttc" id="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4_html_a179fabfc4904fe6ba064116dd27be8d7"><div class="ttname"><a href="structmshadow_1_1expr_1_1BLASEngine_3_01cpu_00_01float_01_4.html#a179fabfc4904fe6ba064116dd27be8d7">mshadow::expr::BLASEngine&lt; cpu, float &gt;::GetT</a></div><div class="ttdeci">static CBLAS_TRANSPOSE GetT(bool t)</div><div class="ttdef"><b>Definition:</b> dot_engine-inl.h:294</div></div>
</div><!-- fragment --></div><!-- contents -->
<!-- start footer part -->
<hr class="footer"/><address class="footer"><small>
Generated on Thu Jan 5 2023 00:58:42 for mxnet by &#160;<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/>
</a> 1.8.13
</small></address>
</body>
</html>