blob: 74e9c4f39b4a012ace54ae2d4b80a97ce5778331 [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"/>
<title>Neural Network Convolution Functions</title>
<title>CMSIS-NN: Neural Network Convolution Functions</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<link href="cmsis.css" rel="stylesheet" type="text/css" />
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<script type="text/javascript" src="printComponentTabs.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
$(document).ready(initResizable);
$(window).load(resizeHeight);
</script>
<link href="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/search.js"></script>
<script type="text/javascript">
$(document).ready(function() { searchBox.OnSelectItem(0); });
</script>
</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: 46px;">
<td id="projectlogo"><img alt="Logo" src="CMSIS_Logo_Final.png"/></td>
<td style="padding-left: 0.5em;">
<div id="projectname">CMSIS-NN
&#160;<span id="projectnumber">Version 1.2.0</span>
</div>
<div id="projectbrief">CMSIS NN Software Library</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<div id="CMSISnav" class="tabs1">
<ul class="tablist">
<script type="text/javascript">
<!--
writeComponentTabs.call(this);
//-->
</script>
</ul>
</div>
<!-- Generated by Doxygen 1.8.6 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<div id="navrow1" class="tabs">
<ul class="tablist">
<li><a href="index.html"><span>Main&#160;Page</span></a></li>
<li><a href="pages.html"><span>Usage&#160;and&#160;Description</span></a></li>
<li><a href="modules.html"><span>Reference</span></a></li>
<li>
<div id="MSearchBox" class="MSearchBoxInactive">
<span class="left">
<img id="MSearchSelect" src="search/mag_sel.png"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
alt=""/>
<input type="text" id="MSearchField" value="Search" accesskey="S"
onfocus="searchBox.OnSearchFieldFocus(true)"
onblur="searchBox.OnSearchFieldFocus(false)"
onkeyup="searchBox.OnSearchFieldChange(event)"/>
</span><span class="right">
<a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
</span>
</div>
</li>
</ul>
</div>
</div><!-- top -->
<div id="side-nav" class="ui-resizable side-nav-resizable">
<div id="nav-tree">
<div id="nav-tree-contents">
<div id="nav-sync" class="sync"></div>
</div>
</div>
<div id="splitbar" style="-moz-user-select:none;"
class="ui-resizable-handle">
</div>
</div>
<script type="text/javascript">
$(document).ready(function(){initNavTree('group__NNConv.html','');});
</script>
<div id="doc-content">
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
<a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(0)"><span class="SelectionMark">&#160;</span>All</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(1)"><span class="SelectionMark">&#160;</span>Data Structures</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(2)"><span class="SelectionMark">&#160;</span>Namespaces</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(3)"><span class="SelectionMark">&#160;</span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(4)"><span class="SelectionMark">&#160;</span>Functions</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(5)"><span class="SelectionMark">&#160;</span>Variables</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(6)"><span class="SelectionMark">&#160;</span>Enumerations</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(7)"><span class="SelectionMark">&#160;</span>Enumerator</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(8)"><span class="SelectionMark">&#160;</span>Macros</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(9)"><span class="SelectionMark">&#160;</span>Groups</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(10)"><span class="SelectionMark">&#160;</span>Pages</a></div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div class="header">
<div class="summary">
<a href="#func-members">Functions</a> </div>
<div class="headertitle">
<div class="title">Neural Network Convolution Functions<div class="ingroups"><a class="el" href="group__groupNN.html">Neural Network Functions</a></div></div> </div>
</div><!--header-->
<div class="contents">
<table class="memberdecls">
<tr class="heading"><td colspan="2"><h2 class="groupheader"><a name="func-members"></a>
Functions</h2></td></tr>
<tr class="memitem:ga110adcfdaab356c750c6270aa5e05f29"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga110adcfdaab356c750c6270aa5e05f29">arm_convolve_1x1_HWC_q7_fast_nonsquare</a> (const q7_t *Im_in, const uint16_t dim_im_in_x, const uint16_t dim_im_in_y, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel_x, const uint16_t dim_kernel_y, const uint16_t padding_x, const uint16_t padding_y, const uint16_t stride_x, const uint16_t stride_y, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out_x, const uint16_t dim_im_out_y, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga110adcfdaab356c750c6270aa5e05f29"><td class="mdescLeft">&#160;</td><td class="mdescRight">Fast Q7 version of 1x1 convolution (non-sqaure shape) <a href="#ga110adcfdaab356c750c6270aa5e05f29">More...</a><br/></td></tr>
<tr class="separator:ga110adcfdaab356c750c6270aa5e05f29"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga55701f213b198084b52eab53097f1f58"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga55701f213b198084b52eab53097f1f58">arm_convolve_HWC_q15_basic</a> (const q15_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q15_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q15_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q15_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga55701f213b198084b52eab53097f1f58"><td class="mdescLeft">&#160;</td><td class="mdescRight">Basic Q15 convolution function. <a href="#ga55701f213b198084b52eab53097f1f58">More...</a><br/></td></tr>
<tr class="separator:ga55701f213b198084b52eab53097f1f58"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga4efb1ccbbaa7dd936961989dcb443f50"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga4efb1ccbbaa7dd936961989dcb443f50">arm_convolve_HWC_q15_fast</a> (const q15_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q15_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q15_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q15_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga4efb1ccbbaa7dd936961989dcb443f50"><td class="mdescLeft">&#160;</td><td class="mdescRight">Fast Q15 convolution function. <a href="#ga4efb1ccbbaa7dd936961989dcb443f50">More...</a><br/></td></tr>
<tr class="separator:ga4efb1ccbbaa7dd936961989dcb443f50"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga614ec3b71eb96e29952ec3f09e7b9c3c"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga614ec3b71eb96e29952ec3f09e7b9c3c">arm_convolve_HWC_q15_fast_nonsquare</a> (const q15_t *Im_in, const uint16_t dim_im_in_x, const uint16_t dim_im_in_y, const uint16_t ch_im_in, const q15_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel_x, const uint16_t dim_kernel_y, const uint16_t padding_x, const uint16_t padding_y, const uint16_t stride_x, const uint16_t stride_y, const q15_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q15_t *Im_out, const uint16_t dim_im_out_x, const uint16_t dim_im_out_y, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga614ec3b71eb96e29952ec3f09e7b9c3c"><td class="mdescLeft">&#160;</td><td class="mdescRight">Fast Q15 convolution function (non-sqaure shape) <a href="#ga614ec3b71eb96e29952ec3f09e7b9c3c">More...</a><br/></td></tr>
<tr class="separator:ga614ec3b71eb96e29952ec3f09e7b9c3c"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga210ae8d8fc1d12ee15b41f1fa6947681"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga210ae8d8fc1d12ee15b41f1fa6947681">arm_convolve_HWC_q7_basic</a> (const q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga210ae8d8fc1d12ee15b41f1fa6947681"><td class="mdescLeft">&#160;</td><td class="mdescRight">Basic Q7 convolution function. <a href="#ga210ae8d8fc1d12ee15b41f1fa6947681">More...</a><br/></td></tr>
<tr class="separator:ga210ae8d8fc1d12ee15b41f1fa6947681"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga4501fa22c0836002aa47ccc313dce252"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga4501fa22c0836002aa47ccc313dce252">arm_convolve_HWC_q7_basic_nonsquare</a> (const q7_t *Im_in, const uint16_t dim_im_in_x, const uint16_t dim_im_in_y, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel_x, const uint16_t dim_kernel_y, const uint16_t padding_x, const uint16_t padding_y, const uint16_t stride_x, const uint16_t stride_y, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out_x, const uint16_t dim_im_out_y, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga4501fa22c0836002aa47ccc313dce252"><td class="mdescLeft">&#160;</td><td class="mdescRight">Basic Q7 convolution function (non-sqaure shape) <a href="#ga4501fa22c0836002aa47ccc313dce252">More...</a><br/></td></tr>
<tr class="separator:ga4501fa22c0836002aa47ccc313dce252"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:gae00d3c1285907d59657369fc98bcc83f"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#gae00d3c1285907d59657369fc98bcc83f">arm_convolve_HWC_q7_fast</a> (const q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:gae00d3c1285907d59657369fc98bcc83f"><td class="mdescLeft">&#160;</td><td class="mdescRight">Fast Q7 convolution function. <a href="#gae00d3c1285907d59657369fc98bcc83f">More...</a><br/></td></tr>
<tr class="separator:gae00d3c1285907d59657369fc98bcc83f"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:gabc6d6b991024e9e5c5cdbd7489de88ef"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#gabc6d6b991024e9e5c5cdbd7489de88ef">arm_convolve_HWC_q7_fast_nonsquare</a> (const q7_t *Im_in, const uint16_t dim_im_in_x, const uint16_t dim_im_in_y, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel_x, const uint16_t dim_kernel_y, const uint16_t padding_x, const uint16_t padding_y, const uint16_t stride_x, const uint16_t stride_y, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out_x, const uint16_t dim_im_out_y, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:gabc6d6b991024e9e5c5cdbd7489de88ef"><td class="mdescLeft">&#160;</td><td class="mdescRight">Fast Q7 convolution function (non-sqaure shape) <a href="#gabc6d6b991024e9e5c5cdbd7489de88ef">More...</a><br/></td></tr>
<tr class="separator:gabc6d6b991024e9e5c5cdbd7489de88ef"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga98f2ead67d7cbdf558b0cd8a3b8fc148"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga98f2ead67d7cbdf558b0cd8a3b8fc148">arm_convolve_HWC_q7_RGB</a> (const q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga98f2ead67d7cbdf558b0cd8a3b8fc148"><td class="mdescLeft">&#160;</td><td class="mdescRight">Q7 convolution function for RGB image. <a href="#ga98f2ead67d7cbdf558b0cd8a3b8fc148">More...</a><br/></td></tr>
<tr class="separator:ga98f2ead67d7cbdf558b0cd8a3b8fc148"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga02a296cb4f6361a70c3ecf1ef1238292"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga02a296cb4f6361a70c3ecf1ef1238292">arm_depthwise_conv_u8_basic_ver1</a> (const uint8_t *input, const uint16_t input_x, const uint16_t input_y, const uint16_t input_ch, const uint8_t *kernel, const uint16_t kernel_x, const uint16_t kernel_y, const int16_t ch_mult, const int16_t pad_x, const int16_t pad_y, const int16_t stride_x, const int16_t stride_y, const int16_t dilation_x, const int16_t dilation_y, const int32_t *bias, const int32_t input_offset, const int32_t filter_offset, const int32_t output_offset, uint8_t *output, const uint16_t output_x, const uint16_t output_y, const int32_t output_activation_min, const int32_t output_activation_max, const int32_t out_shift, const int32_t out_mult)</td></tr>
<tr class="memdesc:ga02a296cb4f6361a70c3ecf1ef1238292"><td class="mdescLeft">&#160;</td><td class="mdescRight">uint8 depthwise convolution function with asymmetric quantization for even number of channel multiplier and input channels. Unless specified otherwise, arguments are mandatory. Both square and non-square inputs are accepted. <a href="#ga02a296cb4f6361a70c3ecf1ef1238292">More...</a><br/></td></tr>
<tr class="separator:ga02a296cb4f6361a70c3ecf1ef1238292"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:gad3d21b3bc6dbd6f3b97d01104349cb0a"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#gad3d21b3bc6dbd6f3b97d01104349cb0a">arm_depthwise_separable_conv_HWC_q7</a> (const q7_t *Im_in, const uint16_t dim_im_in, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel, const uint16_t padding, const uint16_t stride, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:gad3d21b3bc6dbd6f3b97d01104349cb0a"><td class="mdescLeft">&#160;</td><td class="mdescRight">Q7 depthwise separable convolution function. <a href="#gad3d21b3bc6dbd6f3b97d01104349cb0a">More...</a><br/></td></tr>
<tr class="separator:gad3d21b3bc6dbd6f3b97d01104349cb0a"><td class="memSeparator" colspan="2">&#160;</td></tr>
<tr class="memitem:ga32ac508c5467813a84f74f96655dc697"><td class="memItemLeft" align="right" valign="top">arm_status&#160;</td><td class="memItemRight" valign="bottom"><a class="el" href="group__NNConv.html#ga32ac508c5467813a84f74f96655dc697">arm_depthwise_separable_conv_HWC_q7_nonsquare</a> (const q7_t *Im_in, const uint16_t dim_im_in_x, const uint16_t dim_im_in_y, const uint16_t ch_im_in, const q7_t *wt, const uint16_t ch_im_out, const uint16_t dim_kernel_x, const uint16_t dim_kernel_y, const uint16_t padding_x, const uint16_t padding_y, const uint16_t stride_x, const uint16_t stride_y, const q7_t *bias, const uint16_t bias_shift, const uint16_t out_shift, q7_t *Im_out, const uint16_t dim_im_out_x, const uint16_t dim_im_out_y, q15_t *bufferA, q7_t *bufferB)</td></tr>
<tr class="memdesc:ga32ac508c5467813a84f74f96655dc697"><td class="mdescLeft">&#160;</td><td class="mdescRight">Q7 depthwise separable convolution function (non-square shape) <a href="#ga32ac508c5467813a84f74f96655dc697">More...</a><br/></td></tr>
<tr class="separator:ga32ac508c5467813a84f74f96655dc697"><td class="memSeparator" colspan="2">&#160;</td></tr>
</table>
<a name="details" id="details"></a><h2 class="groupheader">Description</h2>
<p>Perform convolution layer</p>
<p>The convolution is implemented in 2 steps: im2col and GEMM</p>
<p>im2col is a process of converting each patch of image data into a column. After im2col, the convolution is computed as matrix-matrix multiplication.</p>
<p>To reduce the memory footprint, the im2col is performed partially. Each iteration, only a few column (i.e., patches) are generated and computed with GEMM kernels similar to CMSIS-DSP arm_mat_mult functions. </p>
<h2 class="groupheader">Function Documentation</h2>
<a class="anchor" id="ga110adcfdaab356c750c6270aa5e05f29"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_1x1_HWC_q7_fast_nonsquare </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_x</td><td>input tensor dimention x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_y</td><td>input tensor dimention y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_x</td><td>filter kernel size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_y</td><td>filter kernel size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_x</td><td>padding size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_y</td><td>padding size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>convolution stride x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>convolution stride y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_x</td><td>output tensor dimension x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_y</td><td>output tensor dimension y </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p>This function is optimized for convolution with 1x1 kernel size (i.e., dim_kernel_x=1 and dim_kernel_y=1). It can be used for the second half of MobileNets [1] after depthwise separable convolution.</p>
<p>This function is the version with full list of optimization tricks, but with some contraints: ch_im_in is multiple of 4 ch_im_out is multiple of 2</p>
<p>[1] MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications <a href="https://arxiv.org/abs/1704.04861">https://arxiv.org/abs/1704.04861</a> </p>
<p>References <a class="el" href="arm__nnfunctions_8h.html#aefe9c7ce9a65060a244b06dffe74c4b3">arm_nn_mat_mult_kernel_q7_q15_reordered()</a>, <a class="el" href="group__nndata__convert.html#gaba8fd446d5f54760b406ee63b25d1aee">arm_q7_to_q15_reordered_no_shift()</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga55701f213b198084b52eab53097f1f58"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q15_basic </td>
<td>(</td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns <code>ARM_MATH_SUCCESS</code></dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p>This basic version is designed to work for any input tensor and weight dimension. </p>
<p>References <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga4efb1ccbbaa7dd936961989dcb443f50"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q15_fast </td>
<td>(</td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p><b>Input dimension constraints:</b></p>
<p>ch_im_in is multiple of 2</p>
<p>ch_im_out is multipe of 2 </p>
<p>References <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga614ec3b71eb96e29952ec3f09e7b9c3c"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q15_fast_nonsquare </td>
<td>(</td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q15_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_x</td><td>input tensor dimention x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_y</td><td>input tensor dimention y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_x</td><td>filter kernel size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_y</td><td>filter kernel size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_x</td><td>padding size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_y</td><td>padding size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>convolution stride x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>convolution stride y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_x</td><td>output tensor dimension x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_y</td><td>output tensor dimension y </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p><b>Input dimension constraints:</b></p>
<p>ch_im_in is multiple of 2</p>
<p>ch_im_out is multipe of 2 </p>
<p>References <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga210ae8d8fc1d12ee15b41f1fa6947681"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q7_basic </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns <code>ARM_MATH_SUCCESS</code></dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p>This basic version is designed to work for any input tensor and weight dimension. </p>
<p>References <a class="el" href="arm__nnfunctions_8h.html#abc4fb258cfe8500ee68e812a293a80a3">arm_nn_mat_mult_kernel_q7_q15()</a>, <a class="el" href="group__nndata__convert.html#gae349de4dba8d253c89d45794ccf05680">arm_q7_to_q15_no_shift()</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga4501fa22c0836002aa47ccc313dce252"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q7_basic_nonsquare </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_x</td><td>input tensor dimention x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_y</td><td>input tensor dimention y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_x</td><td>filter kernel size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_y</td><td>filter kernel size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_x</td><td>padding size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_y</td><td>padding size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>convolution stride x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>convolution stride y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_x</td><td>output tensor dimension x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_y</td><td>output tensor dimension y </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns <code>ARM_MATH_SUCCESS</code> </dd></dl>
<p>References <a class="el" href="arm__nnfunctions_8h.html#abc4fb258cfe8500ee68e812a293a80a3">arm_nn_mat_mult_kernel_q7_q15()</a>, <a class="el" href="group__nndata__convert.html#gae349de4dba8d253c89d45794ccf05680">arm_q7_to_q15_no_shift()</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="gae00d3c1285907d59657369fc98bcc83f"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q7_fast </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p><b>Input dimension constraints:</b></p>
<p>ch_im_in is multiple of 4 ( because of the SIMD32 read and swap )</p>
<p>ch_im_out is multipe of 2 ( bacause 2x2 mat_mult kernel )</p>
<p>The im2col converts the Q7 tensor input into Q15 column, which is stored in bufferA. There is reordering happenning during this im2col process with arm_q7_to_q15_reordered_no_shift. For every four elements, the second and third elements are swapped.</p>
<p>The computation kernel arm_nn_mat_mult_kernel_q7_q15_reordered does the GEMM computation with the reordered columns.</p>
<p>To speed-up the determination of the padding condition, we split the computation into 3x3 parts, i.e., {top, mid, bottom} X {left, mid, right}. This reduces the total number of boundary condition checks and improves the data copying performance. </p>
<p>References <a class="el" href="arm__nnfunctions_8h.html#aefe9c7ce9a65060a244b06dffe74c4b3">arm_nn_mat_mult_kernel_q7_q15_reordered()</a>, <a class="el" href="group__nndata__convert.html#gaba8fd446d5f54760b406ee63b25d1aee">arm_q7_to_q15_reordered_no_shift()</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
<p>Referenced by <a class="el" href="arm__nnexamples__cifar10_8cpp.html#ae66f6b31b5ad750f1fe042a706a4e3d4">main()</a>.</p>
</div>
</div>
<a class="anchor" id="gabc6d6b991024e9e5c5cdbd7489de88ef"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q7_fast_nonsquare </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_x</td><td>input tensor dimention x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_y</td><td>input tensor dimention y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_x</td><td>filter kernel size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_y</td><td>filter kernel size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_x</td><td>padding size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_y</td><td>padding size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>convolution stride x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>convolution stride y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_x</td><td>output tensor dimension x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_y</td><td>output tensor dimension y </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p>This function is the version with full list of optimization tricks, but with some contraints: ch_im_in is multiple of 4 ch_im_out is multiple of 2 </p>
<p>References <a class="el" href="arm__nnfunctions_8h.html#aefe9c7ce9a65060a244b06dffe74c4b3">arm_nn_mat_mult_kernel_q7_q15_reordered()</a>, <a class="el" href="group__nndata__convert.html#gaba8fd446d5f54760b406ee63b25d1aee">arm_q7_to_q15_reordered_no_shift()</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>.</p>
</div>
</div>
<a class="anchor" id="ga98f2ead67d7cbdf558b0cd8a3b8fc148"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_convolve_HWC_q7_RGB </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<p>Q7 version of convolution for RGB image.</p>
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p><b>Input dimension constraints:</b></p>
<p>ch_im_in equals 3</p>
<p>This kernel is written exclusively for convolution with ch_im_in equals 3. This applies on the first layer of CNNs which has input image with RGB format. </p>
<p>References <a class="el" href="arm__nnfunctions_8h.html#abc4fb258cfe8500ee68e812a293a80a3">arm_nn_mat_mult_kernel_q7_q15()</a>, <a class="el" href="unionarm__nnword.html#a9b5e49e4e2c4b7203e07b305386bb2ba">arm_nnword::half_words</a>, <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>, and <a class="el" href="unionarm__nnword.html#a35c7b2ae25e35e0ddcd9ec0a1a6f8d18">arm_nnword::word</a>.</p>
<p>Referenced by <a class="el" href="arm__nnexamples__cifar10_8cpp.html#ae66f6b31b5ad750f1fe042a706a4e3d4">main()</a>.</p>
</div>
</div>
<a class="anchor" id="ga02a296cb4f6361a70c3ecf1ef1238292"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_depthwise_conv_u8_basic_ver1 </td>
<td>(</td>
<td class="paramtype">const uint8_t *&#160;</td>
<td class="paramname"><em>input</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>input_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>input_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>input_ch</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint8_t *&#160;</td>
<td class="paramname"><em>kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>ch_mult</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>pad_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>pad_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>dilation_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int16_t&#160;</td>
<td class="paramname"><em>dilation_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>input_offset</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>filter_offset</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>output_offset</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">uint8_t *&#160;</td>
<td class="paramname"><em>output</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>output_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>output_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>output_activation_min</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>output_activation_max</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const int32_t&#160;</td>
<td class="paramname"><em>out_mult</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<p>uint8 depthwise convolution function with asymmetric quantization for even number of channel multiplier and input channels. Unless specified otherwise, arguments are mandatory.</p>
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">input</td><td>Pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">input_x</td><td>Width of input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">input_y</td><td>Height of input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">input_ch</td><td>Channels in input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">kernel</td><td>Pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">kernel_x</td><td>Width of kernel </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">kernel_y</td><td>Height of kernel </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_mult</td><td>Number of channel multiplier </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">pad_x</td><td>Padding sizes x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">pad_y</td><td>Padding sizes y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>Convolution stride along the width </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>Convolution stride along the height </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dilation_x</td><td>Dilation along width. Not used and intended for future enhancement. </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dilation_y</td><td>Dilation along height. Not used and intended for future enhancement. </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>Pointer to optional bias values. If no bias is availble, NULL is expected </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">input_offset</td><td>Input tensor zero offset </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">filter_offset</td><td>Kernel tensor zero offset </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">output_offset</td><td>Output tensor zero offset </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">output</td><td>Pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">output_x</td><td>Width of output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">output_y</td><td>Height of output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">output_activation_min</td><td>Minimum value to clamp the output to. Range : {0, 255} </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">output_activation_max</td><td>Minimum value to clamp the output to. Range : {0, 255} </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>Amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_mult</td><td>Output multiplier for requantization </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns one of the following <code>ARM_MATH_SIZE_MISMATCH</code> - Not supported dimension of tensors <code>ARM_MATH_SUCCESS</code> - Successful operation <code>ARM_MATH_ARGUMENT_ERROR</code> - Implementation not available</dd></dl>
<p><b> Input constraints</b> ch_mult is multiple of 2 kernel_x is multiple of 2 </p>
<p>References <a class="el" href="arm__nnsupportfunctions_8h.html#ab6dbc2fd53fae3ccdd1d0d70c8d3b491">arm_nn_divide_by_power_of_two()</a>, <a class="el" href="arm__nnsupportfunctions_8h.html#a51721c10f116c9f5b8d9908367989d44">arm_nn_sat_doubling_high_mult()</a>, <a class="el" href="arm__depthwise__conv__u8__basic__ver1_8c.html#aeb98996ffa34fb40d8e91919d2ebdc56">DILATION_X</a>, <a class="el" href="arm__depthwise__conv__u8__basic__ver1_8c.html#a9371ac7b5689f9e74a18a2d548c32033">DILATION_Y</a>, <a class="el" href="arm__nnsupportfunctions_8h.html#a2953f124ae444ebdd2b2a0873ac36b7e">LEFT_SHIFT</a>, and <a class="el" href="arm__nnsupportfunctions_8h.html#a26af54489c1401b91595bf0c92ef87c4">RIGHT_SHIFT</a>.</p>
</div>
</div>
<a class="anchor" id="gad3d21b3bc6dbd6f3b97d01104349cb0a"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_depthwise_separable_conv_HWC_q7 </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in</td><td>input tensor dimention </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel</td><td>filter kernel size </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding</td><td>padding sizes </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride</td><td>convolution stride </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out</td><td>output tensor dimension </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p><b>Buffer size:</b></p>
<p>bufferA size: 2*ch_im_in*dim_kernel*dim_kernel</p>
<p>bufferB size: 0</p>
<p><b>Input dimension constraints:</b></p>
<p>ch_im_in equals ch_im_out</p>
<p>Implementation: There are 3 nested loop here: Inner loop: calculate each output value with MAC instruction over an accumulator Mid loop: loop over different output channel Outer loop: loop over different output (x, y) </p>
<p>References <a class="el" href="unionarm__nnword.html#ac7cff6480a8e29d95f29b73cb1267249">arm_nnword::bytes</a>, <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>, and <a class="el" href="unionarm__nnword.html#a35c7b2ae25e35e0ddcd9ec0a1a6f8d18">arm_nnword::word</a>.</p>
</div>
</div>
<a class="anchor" id="ga32ac508c5467813a84f74f96655dc697"></a>
<div class="memitem">
<div class="memproto">
<table class="memname">
<tr>
<td class="memname">arm_status arm_depthwise_separable_conv_HWC_q7_nonsquare </td>
<td>(</td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>Im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_in_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_in</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>wt</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>ch_im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_kernel_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>padding_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>stride_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const q7_t *&#160;</td>
<td class="paramname"><em>bias</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>bias_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>out_shift</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>Im_out</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_x</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">const uint16_t&#160;</td>
<td class="paramname"><em>dim_im_out_y</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q15_t *&#160;</td>
<td class="paramname"><em>bufferA</em>, </td>
</tr>
<tr>
<td class="paramkey"></td>
<td></td>
<td class="paramtype">q7_t *&#160;</td>
<td class="paramname"><em>bufferB</em>&#160;</td>
</tr>
<tr>
<td></td>
<td>)</td>
<td></td><td></td>
</tr>
</table>
</div><div class="memdoc">
<dl class="params"><dt>Parameters</dt><dd>
<table class="params">
<tr><td class="paramdir">[in]</td><td class="paramname">Im_in</td><td>pointer to input tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_x</td><td>input tensor dimention x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_in_y</td><td>input tensor dimention y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_in</td><td>number of input tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">wt</td><td>pointer to kernel weights </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">ch_im_out</td><td>number of filters, i.e., output tensor channels </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_x</td><td>filter kernel size x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_kernel_y</td><td>filter kernel size y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_x</td><td>padding sizes x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">padding_y</td><td>padding sizes y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_x</td><td>convolution stride x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">stride_y</td><td>convolution stride y </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias</td><td>pointer to bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">bias_shift</td><td>amount of left-shift for bias </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">out_shift</td><td>amount of right-shift for output </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">Im_out</td><td>pointer to output tensor </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_x</td><td>output tensor dimension x </td></tr>
<tr><td class="paramdir">[in]</td><td class="paramname">dim_im_out_y</td><td>output tensor dimension y </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferA</td><td>pointer to buffer space for input </td></tr>
<tr><td class="paramdir">[in,out]</td><td class="paramname">bufferB</td><td>pointer to buffer space for output </td></tr>
</table>
</dd>
</dl>
<dl class="section return"><dt>Returns</dt><dd>The function returns either <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.</dd></dl>
<p>This function is the version with full list of optimization tricks, but with some contraints: ch_im_in is multiple of 2 ch_im_out is multiple of 2 </p>
<p>References <a class="el" href="unionarm__nnword.html#ac7cff6480a8e29d95f29b73cb1267249">arm_nnword::bytes</a>, <a class="el" href="arm__nnsupportfunctions_8h.html#a4cbd428a2b4a4f6b2a6e4219520c7ce0">NN_ROUND</a>, and <a class="el" href="unionarm__nnword.html#a35c7b2ae25e35e0ddcd9ec0a1a6f8d18">arm_nnword::word</a>.</p>
</div>
</div>
</div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="footer">Generated on Wed Jul 10 2019 15:20:50 for CMSIS-NN Version 1.2.0 by Arm Ltd. All rights reserved.
<!--
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.6
-->
</li>
</ul>
</div>
</body>
</html>