Google Git
Sign in
android / platform / external / ComputeLibrary / 975dfe175 / . / documentation / _n_e_math_8inl_source.xhtml
blob: 0b4b09cacd3f019e11d52ea18b90075340f39336 [file] [log] [blame]
<!-- HTML header for doxygen 1.8.15-->
<!-- Remember to use version doxygen 1.8.15 +-->
<!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 1.8.15"/>
<meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
<title>Compute Library: arm_compute/core/NEON/NEMath.inl Source File</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtreedata.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(document).ready(initResizable);
/* @license-end */</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>
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
extensions: ["tex2jax.js"],
jax: ["input/TeX","output/HTML-CSS"],
});
</script><script type="text/javascript" async="async" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
<link href="doxygen.css" rel="stylesheet" type="text/css" />
<link href="stylesheet.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;">
<img alt="Compute Library" src="https://raw.githubusercontent.com/ARM-software/ComputeLibrary/gh-pages/ACL_logo.png" style="max-width: 100%;margin-top: 15px;margin-left: 10px"/>
<td style="padding-left: 0.5em;">
<div id="projectname">
&#160;<span id="projectnumber">19.08</span>
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.15 -->
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
var searchBox = new SearchBox("searchBox", "search",false,'Search');
/* @license-end */
</script>
<script type="text/javascript" src="menudata.js"></script>
<script type="text/javascript" src="menu.js"></script>
<script type="text/javascript">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(function() {
initMenu('',true,false,'search.php','Search');
$(document).ready(function() { init_search(); });
});
/* @license-end */</script>
<div id="main-nav"></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">
/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
$(document).ready(function(){initNavTree('_n_e_math_8inl_source.xhtml','');});
/* @license-end */
</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)">
</div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div class="header">
<div class="headertitle">
<div class="title">NEMath.inl</div> </div>
</div><!--header-->
<div class="contents">
<a href="_n_e_math_8inl.xhtml">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"> * Copyright (c) 2016-2019 ARM Limited.</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00004"></a><span class="lineno"> 4</span>&#160;<span class="comment"> * SPDX-License-Identifier: MIT</span></div><div class="line"><a name="l00005"></a><span class="lineno"> 5</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span>&#160;<span class="comment"> * Permission is hereby granted, free of charge, to any person obtaining a copy</span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span>&#160;<span class="comment"> * of this software and associated documentation files (the &quot;Software&quot;), to</span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span>&#160;<span class="comment"> * deal in the Software without restriction, including without limitation the</span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span>&#160;<span class="comment"> * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or</span></div><div class="line"><a name="l00010"></a><span class="lineno"> 10</span>&#160;<span class="comment"> * sell copies of the Software, and to permit persons to whom the Software is</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span>&#160;<span class="comment"> * furnished to do so, subject to the following conditions:</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span>&#160;<span class="comment"> * The above copyright notice and this permission notice shall be included in all</span></div><div class="line"><a name="l00014"></a><span class="lineno"> 14</span>&#160;<span class="comment"> * copies or substantial portions of the Software.</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span>&#160;<span class="comment"> *</span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span>&#160;<span class="comment"> * THE SOFTWARE IS PROVIDED &quot;AS IS&quot;, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span>&#160;<span class="comment"> * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span>&#160;<span class="comment"> * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span>&#160;<span class="comment"> * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER</span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span>&#160;<span class="comment"> * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span>&#160;<span class="comment"> * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE</span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span>&#160;<span class="comment"> * SOFTWARE.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span>&#160;</div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span>&#160;<span class="preprocessor">#include &lt;cmath&gt;</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span>&#160;</div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span>&#160;<span class="keyword">namespace </span><a class="code" href="namespacearm__compute.xhtml">arm_compute</a></div><div class="line"><a name="l00028"></a><span class="lineno"> 28</span>&#160;{</div><div class="line"><a name="l00030"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#a869945609357fa552d94eb16f7aad4e9"> 30</a></span>&#160;<span class="keyword">const</span> std::array&lt;float32x4_t, 8&gt; <a class="code" href="namespacearm__compute.xhtml#a869945609357fa552d94eb16f7aad4e9">exp_tab</a> =</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; {</div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span>&#160; vdupq_n_f32(1.f),</div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span>&#160; vdupq_n_f32(0.0416598916054f),</div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span>&#160; vdupq_n_f32(0.500000596046f),</div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span>&#160; vdupq_n_f32(0.0014122662833f),</div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span>&#160; vdupq_n_f32(1.00000011921f),</div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span>&#160; vdupq_n_f32(0.00833693705499f),</div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span>&#160; vdupq_n_f32(0.166665703058f),</div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span>&#160; vdupq_n_f32(0.000195780929062f),</div><div class="line"><a name="l00041"></a><span class="lineno"> 41</span>&#160; }</div><div class="line"><a name="l00042"></a><span class="lineno"> 42</span>&#160;};</div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span>&#160;</div><div class="line"><a name="l00045"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#a48fb9cf404e8f7043235bf14105c9793"> 45</a></span>&#160;<span class="keyword">const</span> std::array&lt;float32x4_t, 8&gt; <a class="code" href="namespacearm__compute.xhtml#a48fb9cf404e8f7043235bf14105c9793">log_tab</a> =</div><div class="line"><a name="l00046"></a><span class="lineno"> 46</span>&#160;{</div><div class="line"><a name="l00047"></a><span class="lineno"> 47</span>&#160; {</div><div class="line"><a name="l00048"></a><span class="lineno"> 48</span>&#160; vdupq_n_f32(-2.29561495781f),</div><div class="line"><a name="l00049"></a><span class="lineno"> 49</span>&#160; vdupq_n_f32(-2.47071170807f),</div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span>&#160; vdupq_n_f32(-5.68692588806f),</div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span>&#160; vdupq_n_f32(-0.165253549814f),</div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span>&#160; vdupq_n_f32(5.17591238022f),</div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>&#160; vdupq_n_f32(0.844007015228f),</div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>&#160; vdupq_n_f32(4.58445882797f),</div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span>&#160; vdupq_n_f32(0.0141278216615f),</div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span>&#160; }</div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span>&#160;};</div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span>&#160;</div><div class="line"><a name="l00060"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#ae3e74f28452dea05869826c30b95f54b"> 60</a></span>&#160;constexpr <span class="keywordtype">float</span> <a class="code" href="namespacearm__compute.xhtml#ae3e74f28452dea05869826c30b95f54b">te_sin_coeff2</a> = 0.166666666666f; <span class="comment">// 1/(2*3)</span></div><div class="line"><a name="l00061"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#a9c1bcf45a03708a37132c1e2df5cd2b0"> 61</a></span>&#160;constexpr <span class="keywordtype">float</span> <a class="code" href="namespacearm__compute.xhtml#a9c1bcf45a03708a37132c1e2df5cd2b0">te_sin_coeff3</a> = 0.05f; <span class="comment">// 1/(4*5)</span></div><div class="line"><a name="l00062"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#a542bbd5e844a0a494e3e4078a433d44b"> 62</a></span>&#160;constexpr <span class="keywordtype">float</span> <a class="code" href="namespacearm__compute.xhtml#a542bbd5e844a0a494e3e4078a433d44b">te_sin_coeff4</a> = 0.023809523810f; <span class="comment">// 1/(6*7)</span></div><div class="line"><a name="l00063"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#afe0353e9228b8be37bcdb4a1770f44e3"> 63</a></span>&#160;constexpr <span class="keywordtype">float</span> <a class="code" href="namespacearm__compute.xhtml#afe0353e9228b8be37bcdb4a1770f44e3">te_sin_coeff5</a> = 0.013888888889f; <span class="comment">// 1/(8*9)</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>&#160;</div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span>&#160;<span class="preprocessor">#ifndef DOXYGEN_SKIP_THIS</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#af7a0f3e622aa4bc8c6077d2262a04e2e">vfloorq_f32</a>(float32x4_t val)</div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>&#160;{</div><div class="line"><a name="l00068"></a><span class="lineno"> 68</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_1 = vdupq_n_f32(1.f);</div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span>&#160;</div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span>&#160; <span class="keyword">const</span> int32x4_t z = vcvtq_s32_f32(val);</div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span>&#160; <span class="keyword">const</span> float32x4_t r = vcvtq_f32_s32(z);</div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span>&#160;</div><div class="line"><a name="l00073"></a><span class="lineno"> 73</span>&#160; <span class="keywordflow">return</span> vbslq_f32(vcgtq_f32(r, val), vsubq_f32(r, CONST_1), r);</div><div class="line"><a name="l00074"></a><span class="lineno"> 74</span>&#160;}</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="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a9fb53f220926a32be02f9446aed5bd88">vroundq_rte_f32</a>(float32x4_t val)</div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span>&#160;{</div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span>&#160;<span class="preprocessor">#ifdef __aarch64__</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span>&#160; <span class="keywordflow">return</span> vrndnq_f32(val);</div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>&#160;<span class="preprocessor">#else // __aarch64__</span></div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_HALF_FLOAT = vdupq_n_f32(0.5f);</div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_1_FLOAT = vdupq_n_f32(1.f);</div><div class="line"><a name="l00083"></a><span class="lineno"> 83</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> int32x4_t CONST_1_INT = vdupq_n_s32(1);</div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span>&#160; <span class="keyword">const</span> float32x4_t floor_val = <a class="code" href="namespacearm__compute.xhtml#af7a0f3e622aa4bc8c6077d2262a04e2e">vfloorq_f32</a>(val);</div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span>&#160; <span class="keyword">const</span> float32x4_t diff = vsubq_f32(val, floor_val);</div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span>&#160;</div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span>&#160; <span class="comment">/*</span></div><div class="line"><a name="l00088"></a><span class="lineno"> 88</span>&#160;<span class="comment"> * Select the floor value when (diff&lt;0.5 || (diff==0.5 &amp;&amp; floor_val%2==0).</span></div><div class="line"><a name="l00089"></a><span class="lineno"> 89</span>&#160;<span class="comment"> * This condition is checked by vorrq_u32(vcltq_f32(diff, CONST_HALF_FLOAT) ,vandq_u32(vceqq_f32(diff, CONST_HALF_FLOAT) , vmvnq_u32(vtstq_s32(vandq_s32(vcvtq_s32_f32(floor_val), CONST_1_INT),CONST_1_INT))))</span></div><div class="line"><a name="l00090"></a><span class="lineno"> 90</span>&#160;<span class="comment"> */</span></div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>&#160;</div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span>&#160; <span class="keywordflow">return</span> vbslq_f32(vorrq_u32(vcltq_f32(diff, CONST_HALF_FLOAT), vandq_u32(vceqq_f32(diff, CONST_HALF_FLOAT), vmvnq_u32(vtstq_s32(vandq_s32(vcvtq_s32_f32(floor_val), CONST_1_INT), CONST_1_INT)))),</div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span>&#160; floor_val, vaddq_f32(floor_val, CONST_1_FLOAT));</div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span>&#160;<span class="preprocessor">#endif // __aarch64__</span></div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span>&#160;}</div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span>&#160;</div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span>&#160;<span class="keyword">inline</span> float32x2_t <a class="code" href="namespacearm__compute.xhtml#afc415e7fc5e633a0b85b08b2bba5b5cf">vinvsqrt_f32</a>(float32x2_t x)</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"> 99</span>&#160; float32x2_t sqrt_reciprocal = vrsqrte_f32(x);</div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span>&#160; sqrt_reciprocal = vmul_f32(vrsqrts_f32(vmul_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>&#160; sqrt_reciprocal = vmul_f32(vrsqrts_f32(vmul_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>&#160;</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>&#160; <span class="keywordflow">return</span> sqrt_reciprocal;</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>&#160;}</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"> 106</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#ab8970d7aed07d8649f5e3088455948b8">vinvsqrtq_f32</a>(float32x4_t x)</div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>&#160;{</div><div class="line"><a name="l00108"></a><span class="lineno"> 108</span>&#160; float32x4_t sqrt_reciprocal = vrsqrteq_f32(x);</div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>&#160; sqrt_reciprocal = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00110"></a><span class="lineno"> 110</span>&#160; sqrt_reciprocal = vmulq_f32(vrsqrtsq_f32(vmulq_f32(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span>&#160;</div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>&#160; <span class="keywordflow">return</span> sqrt_reciprocal;</div><div class="line"><a name="l00113"></a><span class="lineno"> 113</span>&#160;}</div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span>&#160;</div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>&#160;<span class="keyword">inline</span> float32x2_t <a class="code" href="namespacearm__compute.xhtml#acc6fd1bcd12ef1141e15401a04d31187">vinv_f32</a>(float32x2_t x)</div><div class="line"><a name="l00116"></a><span class="lineno"> 116</span>&#160;{</div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span>&#160; float32x2_t recip = vrecpe_f32(x);</div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>&#160; recip = vmul_f32(vrecps_f32(x, recip), recip);</div><div class="line"><a name="l00119"></a><span class="lineno"> 119</span>&#160; recip = vmul_f32(vrecps_f32(x, recip), recip);</div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span>&#160; <span class="keywordflow">return</span> recip;</div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>&#160;}</div><div class="line"><a name="l00122"></a><span class="lineno"> 122</span>&#160;</div><div class="line"><a name="l00123"></a><span class="lineno"> 123</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a2e0f2412b4d1302272d1aeb85d21ab25">vinvq_f32</a>(float32x4_t x)</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"> 125</span>&#160; float32x4_t recip = vrecpeq_f32(x);</div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span>&#160; recip = vmulq_f32(vrecpsq_f32(x, recip), recip);</div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>&#160; recip = vmulq_f32(vrecpsq_f32(x, recip), recip);</div><div class="line"><a name="l00128"></a><span class="lineno"> 128</span>&#160; <span class="keywordflow">return</span> recip;</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>&#160;}</div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>&#160;</div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a0c10bb00a86edd8d0cd4c33b494f6cd4">vtaylor_polyq_f32</a>(float32x4_t x, <span class="keyword">const</span> std::array&lt;float32x4_t, 8&gt; &amp;coeffs)</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; float32x4_t <a class="code" href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa7fc56270e7a70fa81a5935b72eacbe29">A</a> = vmlaq_f32(coeffs[0], coeffs[4], x);</div><div class="line"><a name="l00134"></a><span class="lineno"> 134</span>&#160; float32x4_t <a class="code" href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa9d5ed678fe57bcca610140957afab571">B</a> = vmlaq_f32(coeffs[2], coeffs[6], x);</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>&#160; float32x4_t C = vmlaq_f32(coeffs[1], coeffs[5], x);</div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>&#160; float32x4_t D = vmlaq_f32(coeffs[3], coeffs[7], x);</div><div class="line"><a name="l00137"></a><span class="lineno"> 137</span>&#160; float32x4_t x2 = vmulq_f32(x, x);</div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span>&#160; float32x4_t x4 = vmulq_f32(x2, x2);</div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>&#160; float32x4_t res = vmlaq_f32(vmlaq_f32(A, B, x2), vmlaq_f32(C, D, x2), x4);</div><div class="line"><a name="l00140"></a><span class="lineno"> 140</span>&#160; <span class="keywordflow">return</span> res;</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;</div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(float32x4_t x)</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span>&#160;{</div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_LN2 = vdupq_n_f32(0.6931471805f); <span class="comment">// ln(2)</span></div><div class="line"><a name="l00146"></a><span class="lineno"> 146</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_INV_LN2 = vdupq_n_f32(1.4426950408f); <span class="comment">// 1/ln(2)</span></div><div class="line"><a name="l00147"></a><span class="lineno"> 147</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_0 = vdupq_n_f32(0.f);</div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> int32x4_t CONST_NEGATIVE_126 = vdupq_n_s32(-126);</div><div class="line"><a name="l00149"></a><span class="lineno"> 149</span>&#160;</div><div class="line"><a name="l00150"></a><span class="lineno"> 150</span>&#160; <span class="comment">// Perform range reduction [-log(2),log(2)]</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>&#160; int32x4_t m = vcvtq_s32_f32(vmulq_f32(x, CONST_INV_LN2));</div><div class="line"><a name="l00152"></a><span class="lineno"> 152</span>&#160; float32x4_t val = vmlsq_f32(x, vcvtq_f32_s32(m), CONST_LN2);</div><div class="line"><a name="l00153"></a><span class="lineno"> 153</span>&#160;</div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>&#160; <span class="comment">// Polynomial Approximation</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>&#160; float32x4_t poly = <a class="code" href="namespacearm__compute.xhtml#a0c10bb00a86edd8d0cd4c33b494f6cd4">vtaylor_polyq_f32</a>(val, <a class="code" href="namespacearm__compute.xhtml#a869945609357fa552d94eb16f7aad4e9">exp_tab</a>);</div><div class="line"><a name="l00156"></a><span class="lineno"> 156</span>&#160;</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>&#160; <span class="comment">// Reconstruct</span></div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>&#160; poly = vreinterpretq_f32_s32(vqaddq_s32(vreinterpretq_s32_f32(poly), vqshlq_n_s32(m, 23)));</div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>&#160; poly = vbslq_f32(vcltq_s32(m, CONST_NEGATIVE_126), CONST_0, poly);</div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>&#160;</div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>&#160; <span class="keywordflow">return</span> poly;</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>&#160;}</div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span>&#160;</div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(float32x4_t x)</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">static</span> <span class="keyword">const</span> int32x4_t CONST_127 = vdupq_n_s32(127); <span class="comment">// 127</span></div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_LN2 = vdupq_n_f32(0.6931471805f); <span class="comment">// ln(2)</span></div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>&#160;</div><div class="line"><a name="l00169"></a><span class="lineno"> 169</span>&#160; <span class="comment">// Extract exponent</span></div><div class="line"><a name="l00170"></a><span class="lineno"> 170</span>&#160; int32x4_t m = vsubq_s32(vreinterpretq_s32_u32(vshrq_n_u32(vreinterpretq_u32_f32(x), 23)), CONST_127);</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>&#160; float32x4_t val = vreinterpretq_f32_s32(vsubq_s32(vreinterpretq_s32_f32(x), vshlq_n_s32(m, 23)));</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>&#160;</div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>&#160; <span class="comment">// Polynomial Approximation</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>&#160; float32x4_t poly = <a class="code" href="namespacearm__compute.xhtml#a0c10bb00a86edd8d0cd4c33b494f6cd4">vtaylor_polyq_f32</a>(val, <a class="code" href="namespacearm__compute.xhtml#a48fb9cf404e8f7043235bf14105c9793">log_tab</a>);</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>&#160;</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>&#160; <span class="comment">// Reconstruct</span></div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span>&#160; poly = vmlaq_f32(poly, vcvtq_f32_s32(m), CONST_LN2);</div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>&#160;</div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>&#160; <span class="keywordflow">return</span> poly;</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>&#160;}</div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span>&#160;</div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a3e59419da344d02daca9c5b274a1da45">vtanhq_f32</a>(float32x4_t val)</div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span>&#160;{</div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_1 = vdupq_n_f32(1.f);</div><div class="line"><a name="l00185"></a><span class="lineno"> 185</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_2 = vdupq_n_f32(2.f);</div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_MIN_TANH = vdupq_n_f32(-10.f);</div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float32x4_t CONST_MAX_TANH = vdupq_n_f32(10.f);</div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span>&#160;</div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span>&#160; float32x4_t x = vminq_f32(vmaxq_f32(val, CONST_MIN_TANH), CONST_MAX_TANH);</div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span>&#160; float32x4_t exp2x = <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(vmulq_f32(CONST_2, x));</div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span>&#160; float32x4_t num = vsubq_f32(exp2x, CONST_1);</div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span>&#160; float32x4_t den = vaddq_f32(exp2x, CONST_1);</div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span>&#160; float32x4_t tanh = vmulq_f32(num, <a class="code" href="namespacearm__compute.xhtml#a2e0f2412b4d1302272d1aeb85d21ab25">vinvq_f32</a>(den));</div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span>&#160; <span class="keywordflow">return</span> tanh;</div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span>&#160;}</div><div class="line"><a name="l00196"></a><span class="lineno"> 196</span>&#160;</div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#aa6465367a67af2bae4ca46750231dae0">vpowq_f32</a>(float32x4_t val, float32x4_t n)</div><div class="line"><a name="l00198"></a><span class="lineno"> 198</span>&#160;{</div><div class="line"><a name="l00199"></a><span class="lineno"> 199</span>&#160; <span class="keywordflow">return</span> <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(vmulq_f32(n, <a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(val)));</div><div class="line"><a name="l00200"></a><span class="lineno"> 200</span>&#160;}</div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span>&#160;</div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span>&#160;<span class="keyword">inline</span> float32x4_t <a class="code" href="namespacearm__compute.xhtml#a5b6eb5a8c8e8944d781892928f71a698">vsinq_f32</a>(float32x4_t val)</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">const</span> float32x4_t pi_v = vdupq_n_f32(M_PI);</div><div class="line"><a name="l00205"></a><span class="lineno"> 205</span>&#160; <span class="keyword">const</span> float32x4_t pio2_v = vdupq_n_f32(M_PI / 2);</div><div class="line"><a name="l00206"></a><span class="lineno"> 206</span>&#160; <span class="keyword">const</span> float32x4_t ipi_v = vdupq_n_f32(1 / M_PI);</div><div class="line"><a name="l00207"></a><span class="lineno"> 207</span>&#160;</div><div class="line"><a name="l00208"></a><span class="lineno"> 208</span>&#160; <span class="comment">//Find positive or negative</span></div><div class="line"><a name="l00209"></a><span class="lineno"> 209</span>&#160; <span class="keyword">const</span> int32x4_t c_v = vabsq_s32(vcvtq_s32_f32(vmulq_f32(val, ipi_v)));</div><div class="line"><a name="l00210"></a><span class="lineno"> 210</span>&#160; <span class="keyword">const</span> uint32x4_t sign_v = vcleq_f32(val, vdupq_n_f32(0));</div><div class="line"><a name="l00211"></a><span class="lineno"> 211</span>&#160; <span class="keyword">const</span> uint32x4_t odd_v = vandq_u32(vreinterpretq_u32_s32(c_v), vdupq_n_u32(1));</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; uint32x4_t neg_v = veorq_u32(odd_v, sign_v);</div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span>&#160;</div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span>&#160; <span class="comment">//Modulus a - (n * int(a*(1/n)))</span></div><div class="line"><a name="l00216"></a><span class="lineno"> 216</span>&#160; float32x4_t ma = vsubq_f32(vabsq_f32(val), vmulq_f32(pi_v, vcvtq_f32_s32(c_v)));</div><div class="line"><a name="l00217"></a><span class="lineno"> 217</span>&#160; <span class="keyword">const</span> uint32x4_t reb_v = vcgeq_f32(ma, pio2_v);</div><div class="line"><a name="l00218"></a><span class="lineno"> 218</span>&#160;</div><div class="line"><a name="l00219"></a><span class="lineno"> 219</span>&#160; <span class="comment">//Rebase a between 0 and pi/2</span></div><div class="line"><a name="l00220"></a><span class="lineno"> 220</span>&#160; ma = vbslq_f32(reb_v, vsubq_f32(pi_v, ma), ma);</div><div class="line"><a name="l00221"></a><span class="lineno"> 221</span>&#160;</div><div class="line"><a name="l00222"></a><span class="lineno"> 222</span>&#160; <span class="comment">//Taylor series</span></div><div class="line"><a name="l00223"></a><span class="lineno"> 223</span>&#160; <span class="keyword">const</span> float32x4_t ma2 = vmulq_f32(ma, ma);</div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span>&#160;</div><div class="line"><a name="l00225"></a><span class="lineno"> 225</span>&#160; <span class="comment">//2nd elem: x^3 / 3!</span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span>&#160; float32x4_t elem = vmulq_f32(vmulq_f32(ma, ma2), vdupq_n_f32(<a class="code" href="namespacearm__compute.xhtml#ae3e74f28452dea05869826c30b95f54b">te_sin_coeff2</a>));</div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span>&#160; float32x4_t res = vsubq_f32(ma, elem);</div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span>&#160;</div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span>&#160; <span class="comment">//3rd elem: x^5 / 5!</span></div><div class="line"><a name="l00230"></a><span class="lineno"> 230</span>&#160; elem = vmulq_f32(vmulq_f32(elem, ma2), vdupq_n_f32(<a class="code" href="namespacearm__compute.xhtml#a9c1bcf45a03708a37132c1e2df5cd2b0">te_sin_coeff3</a>));</div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span>&#160; res = vaddq_f32(res, elem);</div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span>&#160;</div><div class="line"><a name="l00233"></a><span class="lineno"> 233</span>&#160; <span class="comment">//4th elem: x^7 / 7!float32x2_t vsin_f32(float32x2_t val)</span></div><div class="line"><a name="l00234"></a><span class="lineno"> 234</span>&#160; elem = vmulq_f32(vmulq_f32(elem, ma2), vdupq_n_f32(<a class="code" href="namespacearm__compute.xhtml#a542bbd5e844a0a494e3e4078a433d44b">te_sin_coeff4</a>));</div><div class="line"><a name="l00235"></a><span class="lineno"> 235</span>&#160; res = vsubq_f32(res, elem);</div><div class="line"><a name="l00236"></a><span class="lineno"> 236</span>&#160;</div><div class="line"><a name="l00237"></a><span class="lineno"> 237</span>&#160; <span class="comment">//5th elem: x^9 / 9!</span></div><div class="line"><a name="l00238"></a><span class="lineno"> 238</span>&#160; elem = vmulq_f32(vmulq_f32(elem, ma2), vdupq_n_f32(<a class="code" href="namespacearm__compute.xhtml#afe0353e9228b8be37bcdb4a1770f44e3">te_sin_coeff5</a>));</div><div class="line"><a name="l00239"></a><span class="lineno"> 239</span>&#160; res = vaddq_f32(res, elem);</div><div class="line"><a name="l00240"></a><span class="lineno"> 240</span>&#160;</div><div class="line"><a name="l00241"></a><span class="lineno"> 241</span>&#160; <span class="comment">//Change of sign</span></div><div class="line"><a name="l00242"></a><span class="lineno"> 242</span>&#160; neg_v = vshlq_n_u32(neg_v, 31);</div><div class="line"><a name="l00243"></a><span class="lineno"> 243</span>&#160; res = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(res), neg_v));</div><div class="line"><a name="l00244"></a><span class="lineno"> 244</span>&#160; <span class="keywordflow">return</span> res;</div><div class="line"><a name="l00245"></a><span class="lineno"> 245</span>&#160;}</div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span>&#160;</div><div class="line"><a name="l00247"></a><span class="lineno"> 247</span>&#160;<span class="keyword">inline</span> float32x2_t <a class="code" href="namespacearm__compute.xhtml#aa9648abfe8fa6a9f3dab23ec9f55e2f1">vsin_f32</a>(float32x2_t val)</div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span>&#160;{</div><div class="line"><a name="l00249"></a><span class="lineno"> 249</span>&#160; <span class="keyword">const</span> float32x2_t pi_v = vdup_n_f32(M_PI);</div><div class="line"><a name="l00250"></a><span class="lineno"> 250</span>&#160; <span class="keyword">const</span> float32x2_t pio2_v = vdup_n_f32(M_PI / 2);</div><div class="line"><a name="l00251"></a><span class="lineno"> 251</span>&#160; <span class="keyword">const</span> float32x2_t ipi_v = vdup_n_f32(1 / M_PI);</div><div class="line"><a name="l00252"></a><span class="lineno"> 252</span>&#160;</div><div class="line"><a name="l00253"></a><span class="lineno"> 253</span>&#160; <span class="comment">//Find positive or negative</span></div><div class="line"><a name="l00254"></a><span class="lineno"> 254</span>&#160; <span class="keyword">const</span> int32x2_t c_v = vabs_s32(vcvt_s32_f32(vmul_f32(val, ipi_v)));</div><div class="line"><a name="l00255"></a><span class="lineno"> 255</span>&#160; <span class="keyword">const</span> uint32x2_t sign_v = vcle_f32(val, vdup_n_f32(0));</div><div class="line"><a name="l00256"></a><span class="lineno"> 256</span>&#160; <span class="keyword">const</span> uint32x2_t odd_v = vand_u32(vreinterpret_u32_s32(c_v), vdup_n_u32(1));</div><div class="line"><a name="l00257"></a><span class="lineno"> 257</span>&#160;</div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span>&#160; uint32x2_t neg_v = veor_u32(odd_v, sign_v);</div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span>&#160;</div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span>&#160; <span class="comment">//Modulus a - (n * int(a*(1/n)))</span></div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>&#160; float32x2_t ma = vsub_f32(vabs_f32(val), vmul_f32(pi_v, vcvt_f32_s32(c_v)));</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>&#160; <span class="keyword">const</span> uint32x2_t reb_v = vcge_f32(ma, pio2_v);</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>&#160;</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>&#160; <span class="comment">//Rebase a between 0 and pi/2</span></div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>&#160; ma = vbsl_f32(reb_v, vsub_f32(pi_v, ma), ma);</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>&#160;</div><div class="line"><a name="l00267"></a><span class="lineno"> 267</span>&#160; <span class="comment">//Taylor series</span></div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span>&#160; <span class="keyword">const</span> float32x2_t ma2 = vmul_f32(ma, ma);</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="comment">//2nd elem: x^3 / 3!</span></div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>&#160; float32x2_t elem = vmul_f32(vmul_f32(ma, ma2), vdup_n_f32(<a class="code" href="namespacearm__compute.xhtml#ae3e74f28452dea05869826c30b95f54b">te_sin_coeff2</a>));</div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>&#160; float32x2_t res = vsub_f32(ma, elem);</div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>&#160;</div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>&#160; <span class="comment">//3rd elem: x^5 / 5!</span></div><div class="line"><a name="l00275"></a><span class="lineno"> 275</span>&#160; elem = vmul_f32(vmul_f32(elem, ma2), vdup_n_f32(<a class="code" href="namespacearm__compute.xhtml#a9c1bcf45a03708a37132c1e2df5cd2b0">te_sin_coeff3</a>));</div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span>&#160; res = vadd_f32(res, elem);</div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>&#160;</div><div class="line"><a name="l00278"></a><span class="lineno"> 278</span>&#160; <span class="comment">//4th elem: x^7 / 7!float32x2_t vsin_f32(float32x2_t val)</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span>&#160; elem = vmul_f32(vmul_f32(elem, ma2), vdup_n_f32(<a class="code" href="namespacearm__compute.xhtml#a542bbd5e844a0a494e3e4078a433d44b">te_sin_coeff4</a>));</div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>&#160; res = vsub_f32(res, elem);</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="comment">//5th elem: x^9 / 9!</span></div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span>&#160; elem = vmul_f32(vmul_f32(elem, ma2), vdup_n_f32(<a class="code" href="namespacearm__compute.xhtml#afe0353e9228b8be37bcdb4a1770f44e3">te_sin_coeff5</a>));</div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>&#160; res = vadd_f32(res, elem);</div><div class="line"><a name="l00285"></a><span class="lineno"> 285</span>&#160;</div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span>&#160; <span class="comment">//Change of sign</span></div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>&#160; neg_v = vshl_n_u32(neg_v, 31);</div><div class="line"><a name="l00288"></a><span class="lineno"> 288</span>&#160; res = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(res), neg_v));</div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span>&#160; <span class="keywordflow">return</span> res;</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;</div><div class="line"><a name="l00292"></a><span class="lineno"> 292</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* DOXYGEN_SKIP_THIS */</span><span class="preprocessor"></span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span>&#160;</div><div class="line"><a name="l00294"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#a5e28a3911204ac952601f512c82b48f1"> 294</a></span>&#160;<span class="keyword">inline</span> int32x4_t <a class="code" href="namespacearm__compute.xhtml#a5e28a3911204ac952601f512c82b48f1">rounding_divide_by_pow2</a>(int32x4_t x, <span class="keywordtype">int</span> exponent)</div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>&#160;{</div><div class="line"><a name="l00296"></a><span class="lineno"> 296</span>&#160; <span class="keyword">const</span> int32x4_t shift_vec = vdupq_n_s32(-exponent);</div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span>&#160; <span class="keyword">const</span> int32x4_t fixup = vshrq_n_s32(vandq_s32(x, shift_vec), 31);</div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span>&#160; <span class="keyword">const</span> int32x4_t fixed_up_x = vqaddq_s32(x, fixup);</div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>&#160; <span class="keywordflow">return</span> vrshlq_s32(fixed_up_x, shift_vec);</div><div class="line"><a name="l00300"></a><span class="lineno"> 300</span>&#160;}</div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span>&#160;</div><div class="line"><a name="l00302"></a><span class="lineno"><a class="line" href="namespacearm__compute.xhtml#afe3908fe41d9f2559f4a9eab1997fe61"> 302</a></span>&#160;<span class="keyword">inline</span> int32_t <a class="code" href="namespacearm__compute.xhtml#a5e28a3911204ac952601f512c82b48f1">rounding_divide_by_pow2</a>(int32_t x, <span class="keywordtype">int</span> exponent)</div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>&#160;{</div><div class="line"><a name="l00304"></a><span class="lineno"> 304</span>&#160; <span class="keyword">const</span> int32_t mask = (1 &lt;&lt; exponent) - 1;</div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span>&#160; <span class="keyword">const</span> int32_t <a class="code" href="namespacearm__compute_1_1test_1_1validation_1_1reference.xhtml#a6cae310822a46019ba25a00ddb0f7e62">threshold</a> = (mask &gt;&gt; 1) + (x &lt; 0 ? 1 : 0);</div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span>&#160; <span class="keywordflow">return</span> (x &gt;&gt; exponent) + ((x &amp; mask) &gt; <a class="code" href="namespacearm__compute_1_1test_1_1validation_1_1reference.xhtml#a6cae310822a46019ba25a00ddb0f7e62">threshold</a> ? 1 : 0);</div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>&#160;}</div><div class="line"><a name="l00308"></a><span class="lineno"> 308</span>&#160;</div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span>&#160;<span class="preprocessor">#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC</span></div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span>&#160;</div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>&#160;<span class="preprocessor">#ifndef DOXYGEN_SKIP_THIS</span></div><div class="line"><a name="l00313"></a><span class="lineno"> 313</span>&#160;<span class="keyword">inline</span> float16x8_t vfloorq_f16(float16x8_t val)</div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span>&#160;{</div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span>&#160; <span class="keyword">static</span> <span class="keyword">const</span> float16x8_t CONST_1 = vdupq_n_f16(1.f);</div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>&#160;</div><div class="line"><a name="l00317"></a><span class="lineno"> 317</span>&#160; <span class="keyword">const</span> int16x8_t z = vcvtq_s16_f16(val);</div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span>&#160; <span class="keyword">const</span> float16x8_t r = vcvtq_f16_s16(z);</div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span>&#160;</div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span>&#160; <span class="keywordflow">return</span> vbslq_f16(vcgtq_f16(r, val), vsubq_f16(r, CONST_1), r);</div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span>&#160;}</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="keyword">inline</span> float16x8_t vroundq_rte_f16(float16x8_t val)</div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>&#160;{</div><div class="line"><a name="l00325"></a><span class="lineno"> 325</span>&#160; <span class="keywordflow">return</span> vrndnq_f16(val);</div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span>&#160;}</div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span>&#160;</div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span>&#160;<span class="keyword">inline</span> float16x4_t vinvsqrt_f16(float16x4_t x)</div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>&#160;{</div><div class="line"><a name="l00330"></a><span class="lineno"> 330</span>&#160; float16x4_t sqrt_reciprocal = vrsqrte_f16(x);</div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span>&#160; sqrt_reciprocal = vmul_f16(vrsqrts_f16(vmul_f16(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span>&#160; sqrt_reciprocal = vmul_f16(vrsqrts_f16(vmul_f16(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span>&#160; <span class="keywordflow">return</span> sqrt_reciprocal;</div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>&#160;}</div><div class="line"><a name="l00335"></a><span class="lineno"> 335</span>&#160;</div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>&#160;<span class="keyword">inline</span> float16x8_t vinvsqrtq_f16(float16x8_t x)</div><div class="line"><a name="l00337"></a><span class="lineno"> 337</span>&#160;{</div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>&#160; float16x8_t sqrt_reciprocal = vrsqrteq_f16(x);</div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>&#160; sqrt_reciprocal = vmulq_f16(vrsqrtsq_f16(vmulq_f16(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00340"></a><span class="lineno"> 340</span>&#160; sqrt_reciprocal = vmulq_f16(vrsqrtsq_f16(vmulq_f16(x, sqrt_reciprocal), sqrt_reciprocal), sqrt_reciprocal);</div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>&#160; <span class="keywordflow">return</span> sqrt_reciprocal;</div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>&#160;}</div><div class="line"><a name="l00343"></a><span class="lineno"> 343</span>&#160;</div><div class="line"><a name="l00344"></a><span class="lineno"> 344</span>&#160;<span class="keyword">inline</span> float16x4_t vinv_f16(float16x4_t x)</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; float16x4_t recip = vrecpe_f16(x);</div><div class="line"><a name="l00347"></a><span class="lineno"> 347</span>&#160; recip = vmul_f16(vrecps_f16(x, recip), recip);</div><div class="line"><a name="l00348"></a><span class="lineno"> 348</span>&#160; recip = vmul_f16(vrecps_f16(x, recip), recip);</div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span>&#160; <span class="keywordflow">return</span> recip;</div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>&#160;}</div><div class="line"><a name="l00351"></a><span class="lineno"> 351</span>&#160;</div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>&#160;<span class="keyword">inline</span> float16x8_t vinvq_f16(float16x8_t x)</div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>&#160;{</div><div class="line"><a name="l00354"></a><span class="lineno"> 354</span>&#160; float16x8_t recip = vrecpeq_f16(x);</div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>&#160; recip = vmulq_f16(vrecpsq_f16(x, recip), recip);</div><div class="line"><a name="l00356"></a><span class="lineno"> 356</span>&#160; recip = vmulq_f16(vrecpsq_f16(x, recip), recip);</div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>&#160; <span class="keywordflow">return</span> recip;</div><div class="line"><a name="l00358"></a><span class="lineno"> 358</span>&#160;}</div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>&#160;</div><div class="line"><a name="l00360"></a><span class="lineno"> 360</span>&#160;<span class="keyword">inline</span> float16x8_t vtanhq_f16(float16x8_t val)</div><div class="line"><a name="l00361"></a><span class="lineno"> 361</span>&#160;{</div><div class="line"><a name="l00362"></a><span class="lineno"> 362</span>&#160; <span class="keyword">const</span> float16x8_t CONST_1 = vdupq_n_f16(1.f);</div><div class="line"><a name="l00363"></a><span class="lineno"> 363</span>&#160; <span class="keyword">const</span> float16x8_t CONST_2 = vdupq_n_f16(2.f);</div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>&#160; <span class="keyword">const</span> float16x8_t CONST_MIN_TANH = vdupq_n_f16(-10.f);</div><div class="line"><a name="l00365"></a><span class="lineno"> 365</span>&#160; <span class="keyword">const</span> float16x8_t CONST_MAX_TANH = vdupq_n_f16(10.f);</div><div class="line"><a name="l00366"></a><span class="lineno"> 366</span>&#160;</div><div class="line"><a name="l00367"></a><span class="lineno"> 367</span>&#160; <span class="keyword">const</span> float16x8_t x = vminq_f16(vmaxq_f16(val, CONST_MIN_TANH), CONST_MAX_TANH);</div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>&#160; <span class="keyword">const</span> float16x8_t exp2x = vexpq_f16(vmulq_f16(CONST_2, x));</div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>&#160; <span class="keyword">const</span> float16x8_t num = vsubq_f16(exp2x, CONST_1);</div><div class="line"><a name="l00370"></a><span class="lineno"> 370</span>&#160; <span class="keyword">const</span> float16x8_t den = vaddq_f16(exp2x, CONST_1);</div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>&#160; <span class="keyword">const</span> float16x8_t tanh = vmulq_f16(num, vinvq_f16(den));</div><div class="line"><a name="l00372"></a><span class="lineno"> 372</span>&#160; <span class="keywordflow">return</span> tanh;</div><div class="line"><a name="l00373"></a><span class="lineno"> 373</span>&#160;}</div><div class="line"><a name="l00374"></a><span class="lineno"> 374</span>&#160;</div><div class="line"><a name="l00375"></a><span class="lineno"> 375</span>&#160;<span class="keyword">inline</span> float16x8_t vtaylor_polyq_f16(float16x8_t x, <span class="keyword">const</span> std::array&lt;float16x8_t, 8&gt; &amp;coeffs)</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"> 377</span>&#160; <span class="keyword">const</span> float16x8_t <a class="code" href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa7fc56270e7a70fa81a5935b72eacbe29">A</a> = vaddq_f16(coeffs[0], vmulq_f16(coeffs[4], x));</div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>&#160; <span class="keyword">const</span> float16x8_t <a class="code" href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa9d5ed678fe57bcca610140957afab571">B</a> = vaddq_f16(coeffs[2], vmulq_f16(coeffs[6], x));</div><div class="line"><a name="l00379"></a><span class="lineno"> 379</span>&#160; <span class="keyword">const</span> float16x8_t C = vaddq_f16(coeffs[1], vmulq_f16(coeffs[5], x));</div><div class="line"><a name="l00380"></a><span class="lineno"> 380</span>&#160; <span class="keyword">const</span> float16x8_t D = vaddq_f16(coeffs[3], vmulq_f16(coeffs[7], x));</div><div class="line"><a name="l00381"></a><span class="lineno"> 381</span>&#160; <span class="keyword">const</span> float16x8_t x2 = vmulq_f16(x, x);</div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>&#160; <span class="keyword">const</span> float16x8_t x4 = vmulq_f16(x2, x2);</div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>&#160; <span class="keyword">const</span> float16x8_t res = vaddq_f16(vaddq_f16(A, vmulq_f16(B, x2)), vmulq_f16(vaddq_f16(C, vmulq_f16(D, x2)), x4));</div><div class="line"><a name="l00384"></a><span class="lineno"> 384</span>&#160; <span class="keywordflow">return</span> res;</div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>&#160;}</div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span>&#160;</div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span>&#160;<span class="keyword">inline</span> float16x8_t vexpq_f16(float16x8_t x)</div><div class="line"><a name="l00388"></a><span class="lineno"> 388</span>&#160;{</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span>&#160; <span class="comment">// TODO (COMPMID-1535) : Revisit FP16 approximations</span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span>&#160; <span class="keyword">const</span> float32x4_t x_high = vcvt_f32_f16(vget_high_f16(x));</div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span>&#160; <span class="keyword">const</span> float32x4_t x_low = vcvt_f32_f16(vget_low_f16(x));</div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span>&#160;</div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span>&#160; <span class="keyword">const</span> float16x8_t res = vcvt_high_f16_f32(vcvt_f16_f32(<a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(x_low)), <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(x_high));</div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span>&#160; <span class="keywordflow">return</span> res;</div><div class="line"><a name="l00395"></a><span class="lineno"> 395</span>&#160;}</div><div class="line"><a name="l00396"></a><span class="lineno"> 396</span>&#160;</div><div class="line"><a name="l00397"></a><span class="lineno"> 397</span>&#160;<span class="keyword">inline</span> float16x8_t vlogq_f16(float16x8_t x)</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; <span class="comment">// TODO (COMPMID-1535) : Revisit FP16 approximations</span></div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span>&#160; <span class="keyword">const</span> float32x4_t x_high = vcvt_f32_f16(vget_high_f16(x));</div><div class="line"><a name="l00401"></a><span class="lineno"> 401</span>&#160; <span class="keyword">const</span> float32x4_t x_low = vcvt_f32_f16(vget_low_f16(x));</div><div class="line"><a name="l00402"></a><span class="lineno"> 402</span>&#160;</div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>&#160; <span class="keyword">const</span> float16x8_t res = vcvt_high_f16_f32(vcvt_f16_f32(<a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(x_low)), <a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(x_high));</div><div class="line"><a name="l00404"></a><span class="lineno"> 404</span>&#160; <span class="keywordflow">return</span> res;</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"> 406</span>&#160;</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>&#160;<span class="keyword">inline</span> float16x8_t vpowq_f16(float16x8_t val, float16x8_t n)</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>&#160;{</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>&#160; <span class="comment">// TODO (giaiod01) - COMPMID-1535</span></div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>&#160; float32x4_t n0_f32 = vcvt_f32_f16(vget_low_f16(n));</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span>&#160; float32x4_t n1_f32 = vcvt_f32_f16(vget_high_f16(n));</div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span>&#160; float32x4_t val0_f32 = vcvt_f32_f16(vget_low_f16(val));</div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>&#160; float32x4_t val1_f32 = vcvt_f32_f16(vget_high_f16(val));</div><div class="line"><a name="l00414"></a><span class="lineno"> 414</span>&#160;</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>&#160; float32x4_t res0_f32 = <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(vmulq_f32(n0_f32, <a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(val0_f32)));</div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>&#160; float32x4_t res1_f32 = <a class="code" href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">vexpq_f32</a>(vmulq_f32(n1_f32, <a class="code" href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">vlogq_f32</a>(val1_f32)));</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span>&#160;</div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span>&#160; <span class="keywordflow">return</span> vcombine_f16(vcvt_f16_f32(res0_f32), vcvt_f16_f32(res1_f32));</div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>&#160;}</div><div class="line"><a name="l00420"></a><span class="lineno"> 420</span>&#160;</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>&#160;<span class="keyword">inline</span> float16x8_t vsinq_f16(float16x8_t val)</div><div class="line"><a name="l00422"></a><span class="lineno"> 422</span>&#160;{</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span>&#160; <span class="keyword">const</span> float32x4_t val_high = vcvt_f32_f16(vget_high_f16(val));</div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span>&#160; <span class="keyword">const</span> float32x4_t val_low = vcvt_f32_f16(vget_low_f16(val));</div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>&#160;</div><div class="line"><a name="l00426"></a><span class="lineno"> 426</span>&#160; <span class="keyword">const</span> float32x4_t res_high = <a class="code" href="namespacearm__compute.xhtml#a5b6eb5a8c8e8944d781892928f71a698">vsinq_f32</a>(val_high);</div><div class="line"><a name="l00427"></a><span class="lineno"> 427</span>&#160; <span class="keyword">const</span> float32x4_t res_low = <a class="code" href="namespacearm__compute.xhtml#a5b6eb5a8c8e8944d781892928f71a698">vsinq_f32</a>(val_low);</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span>&#160;</div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>&#160; <span class="keywordflow">return</span> vcombine_f16(vcvt_f16_f32(res_low), vcvt_f16_f32(res_high));</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>&#160;}</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="keyword">inline</span> float16x4_t vsin_f16(float16x4_t val)</div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>&#160;{</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span>&#160; <span class="keyword">const</span> float32x4_t val_f32 = vcvt_f32_f16(val);</div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>&#160; <span class="keyword">const</span> float32x2_t val_high = vget_high_f32(val_f32);</div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span>&#160; <span class="keyword">const</span> float32x2_t val_low = vget_low_f32(val_f32);</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; <span class="keyword">const</span> float32x2_t res_high = <a class="code" href="namespacearm__compute.xhtml#aa9648abfe8fa6a9f3dab23ec9f55e2f1">vsin_f32</a>(val_high);</div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span>&#160; <span class="keyword">const</span> float32x2_t res_low = <a class="code" href="namespacearm__compute.xhtml#aa9648abfe8fa6a9f3dab23ec9f55e2f1">vsin_f32</a>(val_low);</div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>&#160;</div><div class="line"><a name="l00441"></a><span class="lineno"> 441</span>&#160; <span class="keywordflow">return</span> vcvt_f16_f32(vcombine_f32(res_low, res_high));</div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span>&#160;}</div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span>&#160;</div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* DOXYGEN_SKIP_THIS */</span><span class="preprocessor"></span></div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span>&#160;<span class="preprocessor">#endif </span><span class="comment">/* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */</span><span class="preprocessor"></span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span>&#160;} <span class="comment">// namespace arm_compute</span></div><div class="ttc" id="namespacearm__compute_xhtml_aa9648abfe8fa6a9f3dab23ec9f55e2f1"><div class="ttname"><a href="namespacearm__compute.xhtml#aa9648abfe8fa6a9f3dab23ec9f55e2f1">arm_compute::vsin_f32</a></div><div class="ttdeci">float32x2_t vsin_f32(float32x2_t val)</div><div class="ttdoc">Calculate sine.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a3e59419da344d02daca9c5b274a1da45"><div class="ttname"><a href="namespacearm__compute.xhtml#a3e59419da344d02daca9c5b274a1da45">arm_compute::vtanhq_f32</a></div><div class="ttdeci">float32x4_t vtanhq_f32(float32x4_t val)</div><div class="ttdoc">Calculate hyperbolic tangent.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a1ce9b523fd4f3b5bbcadcd796183455aa7fc56270e7a70fa81a5935b72eacbe29"><div class="ttname"><a href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa7fc56270e7a70fa81a5935b72eacbe29">arm_compute::Channel::A</a></div><div class="ttdoc">Alpha channel.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_ab8970d7aed07d8649f5e3088455948b8"><div class="ttname"><a href="namespacearm__compute.xhtml#ab8970d7aed07d8649f5e3088455948b8">arm_compute::vinvsqrtq_f32</a></div><div class="ttdeci">float32x4_t vinvsqrtq_f32(float32x4_t x)</div><div class="ttdoc">Calculate inverse square root.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_afe0353e9228b8be37bcdb4a1770f44e3"><div class="ttname"><a href="namespacearm__compute.xhtml#afe0353e9228b8be37bcdb4a1770f44e3">arm_compute::te_sin_coeff5</a></div><div class="ttdeci">constexpr float te_sin_coeff5</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00063">NEMath.inl:63</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_acc6fd1bcd12ef1141e15401a04d31187"><div class="ttname"><a href="namespacearm__compute.xhtml#acc6fd1bcd12ef1141e15401a04d31187">arm_compute::vinv_f32</a></div><div class="ttdeci">float32x2_t vinv_f32(float32x2_t x)</div><div class="ttdoc">Calculate reciprocal.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a1ce9b523fd4f3b5bbcadcd796183455aa9d5ed678fe57bcca610140957afab571"><div class="ttname"><a href="namespacearm__compute.xhtml#a1ce9b523fd4f3b5bbcadcd796183455aa9d5ed678fe57bcca610140957afab571">arm_compute::Channel::B</a></div><div class="ttdoc">Blue channel.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a0c10bb00a86edd8d0cd4c33b494f6cd4"><div class="ttname"><a href="namespacearm__compute.xhtml#a0c10bb00a86edd8d0cd4c33b494f6cd4">arm_compute::vtaylor_polyq_f32</a></div><div class="ttdeci">float32x4_t vtaylor_polyq_f32(float32x4_t x, const std::array&lt; float32x4_t, 8 &gt; &amp;coeffs)</div><div class="ttdoc">Perform a 7th degree polynomial approximation using Estrin's method.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml"><div class="ttname"><a href="namespacearm__compute.xhtml">arm_compute</a></div><div class="ttdoc">Copyright (c) 2017-2018 ARM Limited.</div><div class="ttdef"><b>Definition:</b> <a href="00__introduction_8dox_source.xhtml#l00024">00_introduction.dox:24</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_af7a0f3e622aa4bc8c6077d2262a04e2e"><div class="ttname"><a href="namespacearm__compute.xhtml#af7a0f3e622aa4bc8c6077d2262a04e2e">arm_compute::vfloorq_f32</a></div><div class="ttdeci">float32x4_t vfloorq_f32(float32x4_t val)</div><div class="ttdoc">Calculate floor of a vector.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a869945609357fa552d94eb16f7aad4e9"><div class="ttname"><a href="namespacearm__compute.xhtml#a869945609357fa552d94eb16f7aad4e9">arm_compute::exp_tab</a></div><div class="ttdeci">const std::array&lt; float32x4_t, 8 &gt; exp_tab</div><div class="ttdoc">Exponent polynomial coefficients.</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00030">NEMath.inl:30</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_aa6465367a67af2bae4ca46750231dae0"><div class="ttname"><a href="namespacearm__compute.xhtml#aa6465367a67af2bae4ca46750231dae0">arm_compute::vpowq_f32</a></div><div class="ttdeci">float32x4_t vpowq_f32(float32x4_t val, float32x4_t n)</div><div class="ttdoc">Calculate n power of a number.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a9c1bcf45a03708a37132c1e2df5cd2b0"><div class="ttname"><a href="namespacearm__compute.xhtml#a9c1bcf45a03708a37132c1e2df5cd2b0">arm_compute::te_sin_coeff3</a></div><div class="ttdeci">constexpr float te_sin_coeff3</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00061">NEMath.inl:61</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a1d090b80b655e7f8fd8977bb93808b04"><div class="ttname"><a href="namespacearm__compute.xhtml#a1d090b80b655e7f8fd8977bb93808b04">arm_compute::vlogq_f32</a></div><div class="ttdeci">float32x4_t vlogq_f32(float32x4_t x)</div><div class="ttdoc">Calculate logarithm.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a542bbd5e844a0a494e3e4078a433d44b"><div class="ttname"><a href="namespacearm__compute.xhtml#a542bbd5e844a0a494e3e4078a433d44b">arm_compute::te_sin_coeff4</a></div><div class="ttdeci">constexpr float te_sin_coeff4</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00062">NEMath.inl:62</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a5e28a3911204ac952601f512c82b48f1"><div class="ttname"><a href="namespacearm__compute.xhtml#a5e28a3911204ac952601f512c82b48f1">arm_compute::rounding_divide_by_pow2</a></div><div class="ttdeci">int32x4_t rounding_divide_by_pow2(int32x4_t x, int exponent)</div><div class="ttdoc">Round to the nearest division by a power-of-two using exponent.</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00294">NEMath.inl:294</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_afc415e7fc5e633a0b85b08b2bba5b5cf"><div class="ttname"><a href="namespacearm__compute.xhtml#afc415e7fc5e633a0b85b08b2bba5b5cf">arm_compute::vinvsqrt_f32</a></div><div class="ttdeci">float32x2_t vinvsqrt_f32(float32x2_t x)</div><div class="ttdoc">Calculate inverse square root.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a5b6eb5a8c8e8944d781892928f71a698"><div class="ttname"><a href="namespacearm__compute.xhtml#a5b6eb5a8c8e8944d781892928f71a698">arm_compute::vsinq_f32</a></div><div class="ttdeci">float32x4_t vsinq_f32(float32x4_t val)</div><div class="ttdoc">Calculate sine.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a9fb53f220926a32be02f9446aed5bd88"><div class="ttname"><a href="namespacearm__compute.xhtml#a9fb53f220926a32be02f9446aed5bd88">arm_compute::vroundq_rte_f32</a></div><div class="ttdeci">float32x4_t vroundq_rte_f32(float32x4_t val)</div><div class="ttdoc">Calculate round value of a vector to nearest with ties to even.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_ab55486f1153a4c5f507d4923b59fc6a0"><div class="ttname"><a href="namespacearm__compute.xhtml#ab55486f1153a4c5f507d4923b59fc6a0">arm_compute::vexpq_f32</a></div><div class="ttdeci">float32x4_t vexpq_f32(float32x4_t x)</div><div class="ttdoc">Calculate exponential.</div></div>
<div class="ttc" id="namespacearm__compute_xhtml_ae3e74f28452dea05869826c30b95f54b"><div class="ttname"><a href="namespacearm__compute.xhtml#ae3e74f28452dea05869826c30b95f54b">arm_compute::te_sin_coeff2</a></div><div class="ttdeci">constexpr float te_sin_coeff2</div><div class="ttdoc">Sin polynomial coefficients.</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00060">NEMath.inl:60</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a2e0f2412b4d1302272d1aeb85d21ab25"><div class="ttname"><a href="namespacearm__compute.xhtml#a2e0f2412b4d1302272d1aeb85d21ab25">arm_compute::vinvq_f32</a></div><div class="ttdeci">float32x4_t vinvq_f32(float32x4_t x)</div><div class="ttdoc">Calculate reciprocal.</div></div>
<div class="ttc" id="namespacearm__compute_1_1test_1_1validation_1_1reference_xhtml_a6cae310822a46019ba25a00ddb0f7e62"><div class="ttname"><a href="namespacearm__compute_1_1test_1_1validation_1_1reference.xhtml#a6cae310822a46019ba25a00ddb0f7e62">arm_compute::test::validation::reference::threshold</a></div><div class="ttdeci">SimpleTensor&lt; T &gt; threshold(const SimpleTensor&lt; T &gt; &amp;src, T threshold, T false_value, T true_value, ThresholdType type, T upper)</div><div class="ttdef"><b>Definition:</b> <a href="reference_2_threshold_8cpp_source.xhtml#l00035">Threshold.cpp:35</a></div></div>
<div class="ttc" id="namespacearm__compute_xhtml_a48fb9cf404e8f7043235bf14105c9793"><div class="ttname"><a href="namespacearm__compute.xhtml#a48fb9cf404e8f7043235bf14105c9793">arm_compute::log_tab</a></div><div class="ttdeci">const std::array&lt; float32x4_t, 8 &gt; log_tab</div><div class="ttdoc">Logarithm polynomial coefficients.</div><div class="ttdef"><b>Definition:</b> <a href="_n_e_math_8inl_source.xhtml#l00045">NEMath.inl:45</a></div></div>
</div><!-- fragment --></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="navelem"><a class="el" href="dir_214608ef36d61442cb2b0c1c4e9a7def.xhtml">arm_compute</a></li><li class="navelem"><a class="el" href="dir_1fb090f0c6070330bfaccc4236d3ca0d.xhtml">core</a></li><li class="navelem"><a class="el" href="dir_315f6cf1ec0b2df3ae747ff8286a19f5.xhtml">NEON</a></li><li class="navelem"><a class="el" href="_n_e_math_8inl.xhtml">NEMath.inl</a></li>
<li class="footer">Generated on Mon Sep 2 2019 11:47:19 for Compute Library by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.15 </li>
</ul>
</div>
</body>
</html>