source: CPL/oasis3-mct/branches/OASIS3-MCT_5.0_branch/doc/doxygen/html/mod__oasis__reprosum_8_f90_source.html @ 6331

Last change on this file since 6331 was 6331, checked in by aclsce, 17 months ago

Moved oasis-mct_5.0 in oasis3-mct/branches directory.

File size: 244.2 KB
Line 
1<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
2<html xmlns="http://www.w3.org/1999/xhtml">
3<head>
4<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
5<meta http-equiv="X-UA-Compatible" content="IE=9"/>
6<meta name="generator" content="Doxygen 1.8.13"/>
7<meta name="viewport" content="width=device-width, initial-scale=1"/>
8<title>Oasis3-MCT: /home/globc2/tcraig/oasis3-mct.trunk/lib/psmile/src/mod_oasis_reprosum.F90 Source File</title>
9<link href="tabs.css" rel="stylesheet" type="text/css"/>
10<script type="text/javascript" src="jquery.js"></script>
11<script type="text/javascript" src="dynsections.js"></script>
12<link href="search/search.css" rel="stylesheet" type="text/css"/>
13<script type="text/javascript" src="search/searchdata.js"></script>
14<script type="text/javascript" src="search/search.js"></script>
15<link href="doxygen.css" rel="stylesheet" type="text/css" />
16</head>
17<body>
18<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
19<div id="titlearea">
20<table cellspacing="0" cellpadding="0">
21 <tbody>
22 <tr style="height: 56px;">
23  <td id="projectalign" style="padding-left: 0.5em;">
24   <div id="projectname">Oasis3-MCT
25   </div>
26  </td>
27 </tr>
28 </tbody>
29</table>
30</div>
31<!-- end header part -->
32<!-- Generated by Doxygen 1.8.13 -->
33<script type="text/javascript">
34var searchBox = new SearchBox("searchBox", "search",false,'Search');
35</script>
36<script type="text/javascript" src="menudata.js"></script>
37<script type="text/javascript" src="menu.js"></script>
38<script type="text/javascript">
39$(function() {
40  initMenu('',true,false,'search.php','Search');
41  $(document).ready(function() { init_search(); });
42});
43</script>
44<div id="main-nav"></div>
45<!-- window showing the filter options -->
46<div id="MSearchSelectWindow"
47     onmouseover="return searchBox.OnSearchSelectShow()"
48     onmouseout="return searchBox.OnSearchSelectHide()"
49     onkeydown="return searchBox.OnSearchSelectKey(event)">
50</div>
51
52<!-- iframe showing the search results (closed by default) -->
53<div id="MSearchResultsWindow">
54<iframe src="javascript:void(0)" frameborder="0" 
55        name="MSearchResults" id="MSearchResults">
56</iframe>
57</div>
58
59<div id="nav-path" class="navpath">
60  <ul>
61<li class="navelem"><a class="el" href="dir_97aefd0d527b934f1d99a682da8fe6a9.html">lib</a></li><li class="navelem"><a class="el" href="dir_6f83a59929ac69cee82ded507f083b9f.html">psmile</a></li><li class="navelem"><a class="el" href="dir_dcfc5ad42d3b5a64023b870fcbdcd859.html">src</a></li>  </ul>
62</div>
63</div><!-- top -->
64<div class="header">
65  <div class="headertitle">
66<div class="title">mod_oasis_reprosum.F90</div>  </div>
67</div><!--header-->
68<div class="contents">
69<a href="mod__oasis__reprosum_8_f90.html">Go to the documentation of this file.</a><div class="fragment"><div class="line"><a name="l00001"></a><span class="lineno">    1</span>&#160;<span class="comment"></span></div><div class="line"><a name="l00002"></a><span class="lineno">    2</span>&#160;<span class="comment">!&gt; OASIS reproducible sum method from P. Worley</span></div><div class="line"><a name="l00003"></a><span class="lineno">    3</span>&#160;</div><div class="line"><a name="l00004"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html">    4</a></span>&#160;<span class="keyword">MODULE</span> <a class="code" href="namespacemod__oasis__reprosum.html">mod_oasis_reprosum</a></div><div class="line"><a name="l00005"></a><span class="lineno">    5</span>&#160;</div><div class="line"><a name="l00006"></a><span class="lineno">    6</span>&#160;<span class="comment">!----------------------------------------------------------------------- </span></div><div class="line"><a name="l00007"></a><span class="lineno">    7</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l00008"></a><span class="lineno">    8</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00009"></a><span class="lineno">    9</span>&#160;<span class="comment">!&gt; Compute reproducible global sums of a set of arrays across an MPI </span></div><div class="line"><a name="l00010"></a><span class="lineno">   10</span>&#160;<span class="comment">!&gt; subcommunicator</span></div><div class="line"><a name="l00011"></a><span class="lineno">   11</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00012"></a><span class="lineno">   12</span>&#160;<span class="comment">! Methods: </span><span class="comment"></span></div><div class="line"><a name="l00013"></a><span class="lineno">   13</span>&#160;<span class="comment">!&gt; Compute using either or both a scalable, reproducible algorithm and a</span></div><div class="line"><a name="l00014"></a><span class="lineno">   14</span>&#160;<span class="comment">!&gt; scalable, nonreproducible algorithm:</span></div><div class="line"><a name="l00015"></a><span class="lineno">   15</span>&#160;<span class="comment">!&gt; * Reproducible (scalable): </span></div><div class="line"><a name="l00016"></a><span class="lineno">   16</span>&#160;<span class="comment">!&gt;    Convert to fixed point (integer vector representation) to enable</span></div><div class="line"><a name="l00017"></a><span class="lineno">   17</span>&#160;<span class="comment">!&gt;    reproducibility when using MPI_Allreduce</span></div><div class="line"><a name="l00018"></a><span class="lineno">   18</span>&#160;<span class="comment">!&gt; * Alternative usually reproducible (scalable):</span></div><div class="line"><a name="l00019"></a><span class="lineno">   19</span>&#160;<span class="comment">!&gt;    Use parallel double-double algorithm due to Helen He and </span></div><div class="line"><a name="l00020"></a><span class="lineno">   20</span>&#160;<span class="comment">!&gt;    Chris Ding, based on David Bailey&#39;s/Don Knuth&#39;s DDPDD algorithm</span></div><div class="line"><a name="l00021"></a><span class="lineno">   21</span>&#160;<span class="comment">!&gt; * Nonreproducible (scalable): </span></div><div class="line"><a name="l00022"></a><span class="lineno">   22</span>&#160;<span class="comment">!&gt;    Floating point and MPI_Allreduce based.</span></div><div class="line"><a name="l00023"></a><span class="lineno">   23</span>&#160;<span class="comment">!&gt; If computing both reproducible and nonreproducible sums, compare </span></div><div class="line"><a name="l00024"></a><span class="lineno">   24</span>&#160;<span class="comment">!&gt; these and report relative difference (if absolute difference</span></div><div class="line"><a name="l00025"></a><span class="lineno">   25</span>&#160;<span class="comment">!&gt; less than sum) or absolute difference back to calling routine.</span></div><div class="line"><a name="l00026"></a><span class="lineno">   26</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00027"></a><span class="lineno">   27</span>&#160;<span class="comment">! Author: P. Worley (based on suggestions from J. White for fixed </span></div><div class="line"><a name="l00028"></a><span class="lineno">   28</span>&#160;<span class="comment">!                    point algorithm and on He/Ding paper for ddpdd </span></div><div class="line"><a name="l00029"></a><span class="lineno">   29</span>&#160;<span class="comment">!                    algorithm)</span></div><div class="line"><a name="l00030"></a><span class="lineno">   30</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00031"></a><span class="lineno">   31</span>&#160;<span class="comment">! Modified by T.Craig for Oasis, May 2017 based on the public version in</span></div><div class="line"><a name="l00032"></a><span class="lineno">   32</span>&#160;<span class="comment">! CESM1.2.</span></div><div class="line"><a name="l00033"></a><span class="lineno">   33</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l00034"></a><span class="lineno">   34</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00035"></a><span class="lineno">   35</span>&#160;</div><div class="line"><a name="l00036"></a><span class="lineno">   36</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00037"></a><span class="lineno">   37</span>&#160;<span class="comment">!- use statements ------------------------------------------------------</span></div><div class="line"><a name="l00038"></a><span class="lineno">   38</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00039"></a><span class="lineno">   39</span>&#160;<span class="preprocessor">#if ( defined noI8 )</span></div><div class="line"><a name="l00040"></a><span class="lineno">   40</span>&#160;<span class="preprocessor"></span>   <span class="comment">! Workaround for when shr_kind_i8 is not supported.</span></div><div class="line"><a name="l00041"></a><span class="lineno">   41</span>&#160;   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__kinds.html">mod_oasis_kinds</a><span class="keywordtype">, only</span>: r8 =&gt; <a class="code" href="namespacemod__oasis__kinds.html#a6129e979d32b55cb020fb76068ce9808">ip_r8_p</a>, i8 =&gt; <a class="code" href="namespacemod__oasis__kinds.html#a1e3c90ad6857ff8675d3e2548d516200">ip_i4_p</a></div><div class="line"><a name="l00042"></a><span class="lineno">   42</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00043"></a><span class="lineno">   43</span>&#160;<span class="preprocessor"></span>   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__kinds.html">mod_oasis_kinds</a><span class="keywordtype">, only</span>: r8 =&gt; <a class="code" href="namespacemod__oasis__kinds.html#a6129e979d32b55cb020fb76068ce9808">ip_r8_p</a>, i8 =&gt; <a class="code" href="namespacemod__oasis__kinds.html#a8dbd6aa926433f02df94bca462f122df">ip_i8_p</a></div><div class="line"><a name="l00044"></a><span class="lineno">   44</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00045"></a><span class="lineno">   45</span>&#160;<span class="preprocessor"></span>   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__data.html">mod_oasis_data</a><span class="keywordtype">,  only</span>: <a class="code" href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">nulprt</a></div><div class="line"><a name="l00046"></a><span class="lineno">   46</span>&#160;   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__sys.html">mod_oasis_sys</a><span class="keywordtype">,   only</span>: <a class="code" href="namespacemod__oasis__sys.html#af1943a0ce0c5765ba8a714379065464f">oasis_abort</a>, <a class="code" href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">estr</a></div><div class="line"><a name="l00047"></a><span class="lineno">   47</span>&#160;   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__mpi.html">mod_oasis_mpi</a><span class="keywordtype">,   only</span>: <a class="code" href="namespacemod__oasis__mpi.html#aaf477847380fbbbc1341520b0b3e66da">oasis_mpi_barrier</a></div><div class="line"><a name="l00048"></a><span class="lineno">   48</span>&#160;   <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__timer.html">mod_oasis_timer</a><span class="keywordtype">, only</span>: <a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>, <a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a></div><div class="line"><a name="l00049"></a><span class="lineno">   49</span>&#160;</div><div class="line"><a name="l00050"></a><span class="lineno">   50</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00051"></a><span class="lineno">   51</span>&#160;<span class="comment">!- module boilerplate --------------------------------------------------</span></div><div class="line"><a name="l00052"></a><span class="lineno">   52</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00053"></a><span class="lineno">   53</span>&#160;   <span class="keywordtype">implicit none</span></div><div class="line"><a name="l00054"></a><span class="lineno">   54</span>&#160;   <span class="keywordtype">private</span></div><div class="line"><a name="l00055"></a><span class="lineno">   55</span>&#160;</div><div class="line"><a name="l00056"></a><span class="lineno">   56</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00057"></a><span class="lineno">   57</span>&#160;<span class="comment">!- include statements --------------------------------------------------</span></div><div class="line"><a name="l00058"></a><span class="lineno">   58</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00059"></a><span class="lineno">   59</span>&#160;<span class="preprocessor">#include &lt;mpif.h&gt;</span></div><div class="line"><a name="l00060"></a><span class="lineno">   60</span>&#160;<span class="preprocessor"></span></div><div class="line"><a name="l00061"></a><span class="lineno">   61</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00062"></a><span class="lineno">   62</span>&#160;<span class="comment">! Public interfaces ----------------------------------------------------</span></div><div class="line"><a name="l00063"></a><span class="lineno">   63</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00064"></a><span class="lineno">   64</span>&#160;   <span class="keywordtype">public</span> :: &amp;</div><div class="line"><a name="l00065"></a><span class="lineno">   65</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">oasis_reprosum_setopts</a>,        &amp;<span class="comment">! set runtime options</span></div><div class="line"><a name="l00066"></a><span class="lineno">   66</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">oasis_reprosum_calc</a>,           &amp;<span class="comment">! calculate distributed sum</span></div><div class="line"><a name="l00067"></a><span class="lineno">   67</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">oasis_reprosum_tolexceeded</a>      <span class="comment">! utility function to check relative </span></div><div class="line"><a name="l00068"></a><span class="lineno">   68</span>&#160;                                      <span class="comment">!  differences against the tolerance</span></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="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00071"></a><span class="lineno">   71</span>&#160;<span class="comment">! Public data ----------------------------------------------------------</span></div><div class="line"><a name="l00072"></a><span class="lineno">   72</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00073"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a89140c250e34a903bc9aa546574988d6">   73</a></span>&#160;   <span class="keywordtype">logical</span>, <span class="keywordtype">public</span>     :: <a class="code" href="namespacemod__oasis__reprosum.html#a89140c250e34a903bc9aa546574988d6">oasis_reprosum_recompute</a> = .false.</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"><a class="line" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">   75</a></span>&#160;   <span class="keywordtype">real(r8)</span>, <span class="keywordtype">public</span>    :: <a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> = -1.0_r8</div><div class="line"><a name="l00076"></a><span class="lineno">   76</span>&#160;</div><div class="line"><a name="l00077"></a><span class="lineno">   77</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00078"></a><span class="lineno">   78</span>&#160;<span class="comment">! Private interfaces ---------------------------------------------------</span></div><div class="line"><a name="l00079"></a><span class="lineno">   79</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00080"></a><span class="lineno">   80</span>&#160;   <span class="keywordtype">private</span> :: &amp;</div><div class="line"><a name="l00081"></a><span class="lineno">   81</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">ddpdd</a>,           &amp;<span class="comment">! double-double sum routine</span></div><div class="line"><a name="l00082"></a><span class="lineno">   82</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">split_indices</a>     <span class="comment">! split indices among OMP threads</span></div><div class="line"><a name="l00083"></a><span class="lineno">   83</span>&#160;</div><div class="line"><a name="l00084"></a><span class="lineno">   84</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00085"></a><span class="lineno">   85</span>&#160;<span class="comment">! Private data ---------------------------------------------------------</span></div><div class="line"><a name="l00086"></a><span class="lineno">   86</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00087"></a><span class="lineno">   87</span>&#160;</div><div class="line"><a name="l00088"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">   88</a></span>&#160;   <span class="keywordtype">logical</span> :: <a class="code" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">repro_sum_use_ddpdd</a> = .false.</div><div class="line"><a name="l00089"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">   89</a></span>&#160;   <span class="keywordtype">logical</span> :: <a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a> = .false.</div><div class="line"><a name="l00090"></a><span class="lineno">   90</span>&#160;</div><div class="line"><a name="l00091"></a><span class="lineno">   91</span>&#160;   <span class="keyword">CONTAINS</span></div><div class="line"><a name="l00092"></a><span class="lineno">   92</span>&#160;</div><div class="line"><a name="l00093"></a><span class="lineno">   93</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l00094"></a><span class="lineno">   94</span>&#160;<span class="comment">!----------------------------------------------------------------------- </span></div><div class="line"><a name="l00095"></a><span class="lineno">   95</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00096"></a><span class="lineno">   96</span>&#160;<span class="comment">!&gt; Set runtime options</span></div><div class="line"><a name="l00097"></a><span class="lineno">   97</span>&#160;<span class="comment">! Author: P. Worley</span></div><div class="line"><a name="l00098"></a><span class="lineno">   98</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00099"></a><span class="lineno">   99</span>&#160;</div><div class="line"><a name="l00100"></a><span class="lineno">  100</span>&#160;<span class="keyword">   subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">oasis_reprosum_setopts</a>(repro_sum_use_ddpdd_in,    &amp;</div><div class="line"><a name="l00101"></a><span class="lineno">  101</span>&#160;                                     repro_sum_rel_diff_max_in, &amp;</div><div class="line"><a name="l00102"></a><span class="lineno">  102</span>&#160;                                     repro_sum_recompute_in,    &amp;</div><div class="line"><a name="l00103"></a><span class="lineno">  103</span>&#160;                                     repro_sum_master,          &amp;</div><div class="line"><a name="l00104"></a><span class="lineno">  104</span>&#160;                                     repro_sum_logunit          )</div><div class="line"><a name="l00105"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">  105</a></span>&#160;</div><div class="line"><a name="l00106"></a><span class="lineno">  106</span>&#160;<span class="comment">!------------------------------Arguments--------------------------------</span></div><div class="line"><a name="l00107"></a><span class="lineno">  107</span>&#160;      <span class="keywordtype">logical</span>, <span class="keywordtype">intent(in)</span>, <span class="keywordtype">optional</span> :: repro_sum_use_ddpdd_in</div><div class="line"><a name="l00108"></a><span class="lineno">  108</span>&#160;      <span class="comment">!&lt; Use DDPDD algorithm instead of fixed precision algorithm</span></div><div class="line"><a name="l00109"></a><span class="lineno">  109</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span>, <span class="keywordtype">optional</span> :: repro_sum_rel_diff_max_in</div><div class="line"><a name="l00110"></a><span class="lineno">  110</span>&#160;      <span class="comment">!&lt; maximum permissible difference between reproducible and </span></div><div class="line"><a name="l00111"></a><span class="lineno">  111</span>&#160;<span class="comment">      !&lt; nonreproducible sums</span></div><div class="line"><a name="l00112"></a><span class="lineno">  112</span>&#160;      <span class="keywordtype">logical</span>, <span class="keywordtype">intent(in)</span>, <span class="keywordtype">optional</span>  :: repro_sum_recompute_in</div><div class="line"><a name="l00113"></a><span class="lineno">  113</span>&#160;      <span class="comment">!&lt; recompute using different algorithm when difference between </span></div><div class="line"><a name="l00114"></a><span class="lineno">  114</span>&#160;<span class="comment">      !&lt; reproducible and nonreproducible sums is too great </span></div><div class="line"><a name="l00115"></a><span class="lineno">  115</span>&#160;      <span class="keywordtype">logical</span>, <span class="keywordtype">intent(in)</span>, <span class="keywordtype">optional</span>  :: repro_sum_master</div><div class="line"><a name="l00116"></a><span class="lineno">  116</span>&#160;      <span class="comment">!&lt; flag indicating whether this process should output</span></div><div class="line"><a name="l00117"></a><span class="lineno">  117</span>&#160;<span class="comment">      !&lt; log messages</span></div><div class="line"><a name="l00118"></a><span class="lineno">  118</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span>, <span class="keywordtype">optional</span>  :: repro_sum_logunit</div><div class="line"><a name="l00119"></a><span class="lineno">  119</span>&#160;      <span class="comment">!&lt; unit number for log messages</span></div><div class="line"><a name="l00120"></a><span class="lineno">  120</span>&#160;<span class="comment">!---------------------------Local Workspace-----------------------------</span></div><div class="line"><a name="l00121"></a><span class="lineno">  121</span>&#160;      <span class="keywordtype">integer</span> llogunit           <span class="comment">! unit number for log messages</span></div><div class="line"><a name="l00122"></a><span class="lineno">  122</span>&#160;      <span class="keywordtype">logical</span> master             <span class="comment">! local master?</span></div><div class="line"><a name="l00123"></a><span class="lineno">  123</span>&#160;      <span class="keywordtype">logical</span>,<span class="keywordtype">save</span> :: firstcall = .true.  <span class="comment">! first call</span></div><div class="line"><a name="l00124"></a><span class="lineno">  124</span>&#160;      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_setopts)&#39;</span></div><div class="line"><a name="l00125"></a><span class="lineno">  125</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00126"></a><span class="lineno">  126</span>&#160;</div><div class="line"><a name="l00127"></a><span class="lineno">  127</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_master) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00128"></a><span class="lineno">  128</span>&#160;         master = repro_sum_master</div><div class="line"><a name="l00129"></a><span class="lineno">  129</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l00130"></a><span class="lineno">  130</span>&#160;         master = .false.</div><div class="line"><a name="l00131"></a><span class="lineno">  131</span>&#160;<span class="keywordflow">      endif</span></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;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_logunit) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00134"></a><span class="lineno">  134</span>&#160;         llogunit = repro_sum_logunit</div><div class="line"><a name="l00135"></a><span class="lineno">  135</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l00136"></a><span class="lineno">  136</span>&#160;         llogunit = <a class="code" href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">nulprt</a></div><div class="line"><a name="l00137"></a><span class="lineno">  137</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00138"></a><span class="lineno">  138</span>&#160;</div><div class="line"><a name="l00139"></a><span class="lineno">  139</span>&#160;      <span class="keywordflow">if</span> (.not. firstcall) <span class="keywordflow">then</span></div><div class="line"><a name="l00140"></a><span class="lineno">  140</span>&#160;         <span class="keyword">write</span>(llogunit,*) subname,<a class="code" href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">estr</a>,<span class="stringliteral">&#39; can only be called once&#39;</span></div><div class="line"><a name="l00141"></a><span class="lineno">  141</span>&#160;         <span class="keyword">call </span><a class="code" href="namespacemod__oasis__sys.html#af1943a0ce0c5765ba8a714379065464f">oasis_abort</a>(file=__file__,line=__line__)</div><div class="line"><a name="l00142"></a><span class="lineno">  142</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00143"></a><span class="lineno">  143</span>&#160;      firstcall = .false.</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="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_use_ddpdd_in) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00146"></a><span class="lineno">  146</span>&#160;         <a class="code" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">repro_sum_use_ddpdd</a> = repro_sum_use_ddpdd_in</div><div class="line"><a name="l00147"></a><span class="lineno">  147</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00148"></a><span class="lineno">  148</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_rel_diff_max_in) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00149"></a><span class="lineno">  149</span>&#160;         <a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> = repro_sum_rel_diff_max_in</div><div class="line"><a name="l00150"></a><span class="lineno">  150</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00151"></a><span class="lineno">  151</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_recompute_in) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00152"></a><span class="lineno">  152</span>&#160;         <a class="code" href="namespacemod__oasis__reprosum.html#a89140c250e34a903bc9aa546574988d6">oasis_reprosum_recompute</a> = repro_sum_recompute_in</div><div class="line"><a name="l00153"></a><span class="lineno">  153</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00154"></a><span class="lineno">  154</span>&#160;      <span class="keywordflow">if</span> (master) <span class="keywordflow">then</span></div><div class="line"><a name="l00155"></a><span class="lineno">  155</span>&#160;         <span class="keywordflow">if</span> ( <a class="code" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">repro_sum_use_ddpdd</a> ) <span class="keywordflow">then</span></div><div class="line"><a name="l00156"></a><span class="lineno">  156</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00157"></a><span class="lineno">  157</span>&#160;              <span class="stringliteral">&#39;Using double-double-based (scalable) usually reproducible &#39;</span>, &amp;</div><div class="line"><a name="l00158"></a><span class="lineno">  158</span>&#160;              <span class="stringliteral">&#39;distributed sum algorithm&#39;</span></div><div class="line"><a name="l00159"></a><span class="lineno">  159</span>&#160;         <span class="keywordflow">else</span></div><div class="line"><a name="l00160"></a><span class="lineno">  160</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00161"></a><span class="lineno">  161</span>&#160;              <span class="stringliteral">&#39;Using fixed-point-based (scalable) reproducible &#39;</span>, &amp;</div><div class="line"><a name="l00162"></a><span class="lineno">  162</span>&#160;              <span class="stringliteral">&#39;distributed sum algorithm&#39;</span></div><div class="line"><a name="l00163"></a><span class="lineno">  163</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l00164"></a><span class="lineno">  164</span>&#160;</div><div class="line"><a name="l00165"></a><span class="lineno">  165</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> &gt;= 0._r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00166"></a><span class="lineno">  166</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00167"></a><span class="lineno">  167</span>&#160;              <span class="stringliteral">&#39; with a maximum relative error tolerance of &#39;</span>, &amp;</div><div class="line"><a name="l00168"></a><span class="lineno">  168</span>&#160;              <a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a></div><div class="line"><a name="l00169"></a><span class="lineno">  169</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a89140c250e34a903bc9aa546574988d6">oasis_reprosum_recompute</a>) <span class="keywordflow">then</span></div><div class="line"><a name="l00170"></a><span class="lineno">  170</span>&#160;               <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00171"></a><span class="lineno">  171</span>&#160;                 <span class="stringliteral">&#39;If tolerance exceeded, sum is recomputed using &#39;</span>, &amp;</div><div class="line"><a name="l00172"></a><span class="lineno">  172</span>&#160;                 <span class="stringliteral">&#39;a serial algorithm.&#39;</span></div><div class="line"><a name="l00173"></a><span class="lineno">  173</span>&#160;            <span class="keywordflow">else</span></div><div class="line"><a name="l00174"></a><span class="lineno">  174</span>&#160;               <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00175"></a><span class="lineno">  175</span>&#160;                 <span class="stringliteral">&#39;If tolerance exceeded, fixed-precision is sum used &#39;</span>, &amp;</div><div class="line"><a name="l00176"></a><span class="lineno">  176</span>&#160;                 <span class="stringliteral">&#39;but a warning is output.&#39;</span></div><div class="line"><a name="l00177"></a><span class="lineno">  177</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00178"></a><span class="lineno">  178</span>&#160;         <span class="keywordflow">else</span></div><div class="line"><a name="l00179"></a><span class="lineno">  179</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname, &amp;</div><div class="line"><a name="l00180"></a><span class="lineno">  180</span>&#160;              <span class="stringliteral">&#39;and not comparing with floating point algorithms.&#39;</span></div><div class="line"><a name="l00181"></a><span class="lineno">  181</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l00182"></a><span class="lineno">  182</span>&#160;</div><div class="line"><a name="l00183"></a><span class="lineno">  183</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00184"></a><span class="lineno">  184</span>&#160;<span class="keyword">   end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">oasis_reprosum_setopts</a></div><div class="line"><a name="l00185"></a><span class="lineno">  185</span>&#160;</div><div class="line"><a name="l00186"></a><span class="lineno">  186</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l00187"></a><span class="lineno">  187</span>&#160;</div><div class="line"><a name="l00188"></a><span class="lineno">  188</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l00189"></a><span class="lineno">  189</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00190"></a><span class="lineno">  190</span>&#160;<span class="comment">!&gt; Compute the global sum of each field in &quot;arr&quot; using the indicated </span></div><div class="line"><a name="l00191"></a><span class="lineno">  191</span>&#160;<span class="comment">!&gt; communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l00192"></a><span class="lineno">  192</span>&#160;<span class="comment">!&gt; on a fixed point algorithm. An alternative is to use an &quot;almost</span></div><div class="line"><a name="l00193"></a><span class="lineno">  193</span>&#160;<span class="comment">!&gt; always reproducible&quot; floating point algorithm.</span></div><div class="line"><a name="l00194"></a><span class="lineno">  194</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00195"></a><span class="lineno">  195</span>&#160;<span class="comment">! The accuracy of the fixed point algorithm is controlled by the</span></div><div class="line"><a name="l00196"></a><span class="lineno">  196</span>&#160;<span class="comment">! number of &quot;levels&quot; of integer expansion. The algorithm will calculate</span></div><div class="line"><a name="l00197"></a><span class="lineno">  197</span>&#160;<span class="comment">! the number of levels that is required for the sum to be essentially</span></div><div class="line"><a name="l00198"></a><span class="lineno">  198</span>&#160;<span class="comment">! exact. The optional parameter arr_max_levels can be used to override</span></div><div class="line"><a name="l00199"></a><span class="lineno">  199</span>&#160;<span class="comment">! the calculated value. The optional parameter arr_max_levels_out can be </span></div><div class="line"><a name="l00200"></a><span class="lineno">  200</span>&#160;<span class="comment">! used to return the values used.</span></div><div class="line"><a name="l00201"></a><span class="lineno">  201</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00202"></a><span class="lineno">  202</span>&#160;<span class="comment">! The algorithm also requires an upper bound on </span></div><div class="line"><a name="l00203"></a><span class="lineno">  203</span>&#160;<span class="comment">! the maximum summand (in absolute value) for each field, and will </span></div><div class="line"><a name="l00204"></a><span class="lineno">  204</span>&#160;<span class="comment">! calculate this internally. However, if the optional parameters</span></div><div class="line"><a name="l00205"></a><span class="lineno">  205</span>&#160;<span class="comment">! arr_max_levels and arr_gbl_max are both set, then the algorithm will</span></div><div class="line"><a name="l00206"></a><span class="lineno">  206</span>&#160;<span class="comment">! use the values in arr_gbl_max for the upper bounds instead. If these</span></div><div class="line"><a name="l00207"></a><span class="lineno">  207</span>&#160;<span class="comment">! are not upper bounds, or if the upper bounds are not tight enough</span></div><div class="line"><a name="l00208"></a><span class="lineno">  208</span>&#160;<span class="comment">! to achieve the requisite accuracy, and if the optional parameter</span></div><div class="line"><a name="l00209"></a><span class="lineno">  209</span>&#160;<span class="comment">! repro_sum_validate is NOT set to .false., the algorithm will repeat the</span></div><div class="line"><a name="l00210"></a><span class="lineno">  210</span>&#160;<span class="comment">! computation with appropriate upper bounds. If only arr_gbl_max is present, </span></div><div class="line"><a name="l00211"></a><span class="lineno">  211</span>&#160;<span class="comment">! then the maxima are computed internally (and the specified values are</span></div><div class="line"><a name="l00212"></a><span class="lineno">  212</span>&#160;<span class="comment">! ignored). The optional parameter arr_gbl_max_out can be </span></div><div class="line"><a name="l00213"></a><span class="lineno">  213</span>&#160;<span class="comment">! used to return the values used.</span></div><div class="line"><a name="l00214"></a><span class="lineno">  214</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00215"></a><span class="lineno">  215</span>&#160;<span class="comment">! Finally, the algorithm requires an upper bound on the number of</span></div><div class="line"><a name="l00216"></a><span class="lineno">  216</span>&#160;<span class="comment">! local summands across all processes. This will be calculated internally, </span></div><div class="line"><a name="l00217"></a><span class="lineno">  217</span>&#160;<span class="comment">! using an MPI collective, but the value in the optional argument </span></div><div class="line"><a name="l00218"></a><span class="lineno">  218</span>&#160;<span class="comment">! gbl_max_nsummands will be used instead if (1) it is present, (2)</span></div><div class="line"><a name="l00219"></a><span class="lineno">  219</span>&#160;<span class="comment">! it is &gt; 0, and (3) the maximum value and required number of levels </span></div><div class="line"><a name="l00220"></a><span class="lineno">  220</span>&#160;<span class="comment">! are also specified. (If the maximum value is calculated, the same </span></div><div class="line"><a name="l00221"></a><span class="lineno">  221</span>&#160;<span class="comment">! MPI collective is used to determine the maximum number of local </span></div><div class="line"><a name="l00222"></a><span class="lineno">  222</span>&#160;<span class="comment">! summands.) The accuracy of the user-specified value is not checked. </span></div><div class="line"><a name="l00223"></a><span class="lineno">  223</span>&#160;<span class="comment">! However, if set to &lt; 1, the value will instead be calculated. If the </span></div><div class="line"><a name="l00224"></a><span class="lineno">  224</span>&#160;<span class="comment">! optional parameter gbl_max_nsummands_out is present, then the value </span></div><div class="line"><a name="l00225"></a><span class="lineno">  225</span>&#160;<span class="comment">! used (gbl_max_nsummands if &gt;= 1; calculated otherwise) will be </span></div><div class="line"><a name="l00226"></a><span class="lineno">  226</span>&#160;<span class="comment">! returned.</span></div><div class="line"><a name="l00227"></a><span class="lineno">  227</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00228"></a><span class="lineno">  228</span>&#160;<span class="comment">! If requested (by setting oasis_reprosum_reldiffmax &gt;= 0.0 and passing in</span></div><div class="line"><a name="l00229"></a><span class="lineno">  229</span>&#160;<span class="comment">! the optional rel_diff parameter), results are compared with a </span></div><div class="line"><a name="l00230"></a><span class="lineno">  230</span>&#160;<span class="comment">! nonreproducible floating point algorithm. </span></div><div class="line"><a name="l00231"></a><span class="lineno">  231</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00232"></a><span class="lineno">  232</span>&#160;<span class="comment">! Note that the cost of the algorithm is not strongly correlated with </span></div><div class="line"><a name="l00233"></a><span class="lineno">  233</span>&#160;<span class="comment">! the number of levels, which primarily shows up as a (modest) increase</span></div><div class="line"><a name="l00234"></a><span class="lineno">  234</span>&#160;<span class="comment">! in cost of the MPI_Allreduce as a function of vector length. Rather the </span></div><div class="line"><a name="l00235"></a><span class="lineno">  235</span>&#160;<span class="comment">! cost is more a function of (a) the number of integers required to </span></div><div class="line"><a name="l00236"></a><span class="lineno">  236</span>&#160;<span class="comment">! represent an individual summand and (b) the number of MPI_Allreduce</span></div><div class="line"><a name="l00237"></a><span class="lineno">  237</span>&#160;<span class="comment">! calls. The number of integers required to represent an individual </span></div><div class="line"><a name="l00238"></a><span class="lineno">  238</span>&#160;<span class="comment">! summand is 1 or 2 when using 8-byte integers for 8-byte real summands </span></div><div class="line"><a name="l00239"></a><span class="lineno">  239</span>&#160;<span class="comment">! when the number of local summands is not too large. As the number of </span></div><div class="line"><a name="l00240"></a><span class="lineno">  240</span>&#160;<span class="comment">! local summands increases, the number of integers required increases. </span></div><div class="line"><a name="l00241"></a><span class="lineno">  241</span>&#160;<span class="comment">! The number of MPI_Allreduce calls is either 2 (specifying nothing) or</span></div><div class="line"><a name="l00242"></a><span class="lineno">  242</span>&#160;<span class="comment">! 1 (specifying gbl_max_nsummands, arr_max_levels, and arr_gbl_max </span></div><div class="line"><a name="l00243"></a><span class="lineno">  243</span>&#160;<span class="comment">! correctly). When specifying arr_max_levels and arr_gbl_max </span></div><div class="line"><a name="l00244"></a><span class="lineno">  244</span>&#160;<span class="comment">! incorrectly, 3 or 4 MPI_Allreduce calls will be required.</span></div><div class="line"><a name="l00245"></a><span class="lineno">  245</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l00246"></a><span class="lineno">  246</span>&#160;<span class="comment">! The alternative algorithm is a minor modification of a parallel</span></div><div class="line"><a name="l00247"></a><span class="lineno">  247</span>&#160;<span class="comment">! implementation of David Bailey&#39;s routine DDPDD by Helen He</span></div><div class="line"><a name="l00248"></a><span class="lineno">  248</span>&#160;<span class="comment">! and Chris Ding. Bailey uses the Knuth trick to implement quadruple</span></div><div class="line"><a name="l00249"></a><span class="lineno">  249</span>&#160;<span class="comment">! precision summation of double precision values with 10 double </span></div><div class="line"><a name="l00250"></a><span class="lineno">  250</span>&#160;<span class="comment">! precision operations. The advantage of this algorithm is that </span></div><div class="line"><a name="l00251"></a><span class="lineno">  251</span>&#160;<span class="comment">! it requires a single MPI_Allreduce and is less expensive per summand</span></div><div class="line"><a name="l00252"></a><span class="lineno">  252</span>&#160;<span class="comment">! than is the fixed precision algorithm. The disadvantage is that it </span></div><div class="line"><a name="l00253"></a><span class="lineno">  253</span>&#160;<span class="comment">! is not guaranteed to be reproducible (though it is reproducible </span></div><div class="line"><a name="l00254"></a><span class="lineno">  254</span>&#160;<span class="comment">! much more often than is the standard algorithm). This alternative </span></div><div class="line"><a name="l00255"></a><span class="lineno">  255</span>&#160;<span class="comment">! is used when the optional parameter ddpdd_sum is set to .true. It is </span></div><div class="line"><a name="l00256"></a><span class="lineno">  256</span>&#160;<span class="comment">! also used if the fixed precision algorithm radix assumption does not </span></div><div class="line"><a name="l00257"></a><span class="lineno">  257</span>&#160;<span class="comment">! hold. </span></div><div class="line"><a name="l00258"></a><span class="lineno">  258</span>&#160;</div><div class="line"><a name="l00259"></a><span class="lineno">  259</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00260"></a><span class="lineno">  260</span>&#160;<span class="keyword">   subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">oasis_reprosum_calc</a> (arr, arr_gsum, nsummands, dsummands,     &amp;</div><div class="line"><a name="l00261"></a><span class="lineno">  261</span>&#160;                                   nflds, ddpdd_sum,                        &amp;</div><div class="line"><a name="l00262"></a><span class="lineno">  262</span>&#160;                                   arr_gbl_max, arr_gbl_max_out,            &amp;</div><div class="line"><a name="l00263"></a><span class="lineno">  263</span>&#160;                                   arr_max_levels, arr_max_levels_out,      &amp;</div><div class="line"><a name="l00264"></a><span class="lineno">  264</span>&#160;                                   gbl_max_nsummands, gbl_max_nsummands_out,&amp;</div><div class="line"><a name="l00265"></a><span class="lineno">  265</span>&#160;                                   gbl_count, repro_sum_validate,           &amp;</div><div class="line"><a name="l00266"></a><span class="lineno">  266</span>&#160;                                   repro_sum_stats, rel_diff, commid        )</div><div class="line"><a name="l00267"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">  267</a></span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00268"></a><span class="lineno">  268</span>&#160;</div><div class="line"><a name="l00269"></a><span class="lineno">  269</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l00270"></a><span class="lineno">  270</span>&#160;</div><div class="line"><a name="l00271"></a><span class="lineno">  271</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nsummands<span class="comment">  !&lt; number of local summands</span></div><div class="line"><a name="l00272"></a><span class="lineno">  272</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: dsummands<span class="comment">  !&lt; declared first dimension</span></div><div class="line"><a name="l00273"></a><span class="lineno">  273</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nflds<span class="comment">      !&lt; number of fields</span></div><div class="line"><a name="l00274"></a><span class="lineno">  274</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span> :: arr(dsummands,nflds) </div><div class="line"><a name="l00275"></a><span class="lineno">  275</span>&#160;                                         <span class="comment">!&lt; input array</span></div><div class="line"><a name="l00276"></a><span class="lineno">  276</span>&#160;</div><div class="line"><a name="l00277"></a><span class="lineno">  277</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>:: arr_gsum(nflds)      </div><div class="line"><a name="l00278"></a><span class="lineno">  278</span>&#160;                                         <span class="comment">!&lt; global means</span></div><div class="line"><a name="l00279"></a><span class="lineno">  279</span>&#160;</div><div class="line"><a name="l00280"></a><span class="lineno">  280</span>&#160;      <span class="keywordtype">logical</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: ddpdd_sum</div><div class="line"><a name="l00281"></a><span class="lineno">  281</span>&#160;                                         <span class="comment">!&lt; use ddpdd algorithm instead</span></div><div class="line"><a name="l00282"></a><span class="lineno">  282</span>&#160;<span class="comment">                                         !&lt; of fixed precision algorithm</span></div><div class="line"><a name="l00283"></a><span class="lineno">  283</span>&#160;</div><div class="line"><a name="l00284"></a><span class="lineno">  284</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: arr_gbl_max(nflds)</div><div class="line"><a name="l00285"></a><span class="lineno">  285</span>&#160;                                         <span class="comment">!&lt; upper bound on max(abs(arr))</span></div><div class="line"><a name="l00286"></a><span class="lineno">  286</span>&#160;</div><div class="line"><a name="l00287"></a><span class="lineno">  287</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>,   <span class="keywordtype">optional</span> :: arr_gbl_max_out(nflds)</div><div class="line"><a name="l00288"></a><span class="lineno">  288</span>&#160;                                         <span class="comment">!&lt; calculated upper bound on </span></div><div class="line"><a name="l00289"></a><span class="lineno">  289</span>&#160;<span class="comment">                                         !&lt; max(abs(arr))</span></div><div class="line"><a name="l00290"></a><span class="lineno">  290</span>&#160;</div><div class="line"><a name="l00291"></a><span class="lineno">  291</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: arr_max_levels(nflds)</div><div class="line"><a name="l00292"></a><span class="lineno">  292</span>&#160;                                         <span class="comment">!&lt; maximum number of levels of</span></div><div class="line"><a name="l00293"></a><span class="lineno">  293</span>&#160;<span class="comment">                                         !&lt; integer expansion to use</span></div><div class="line"><a name="l00294"></a><span class="lineno">  294</span>&#160;</div><div class="line"><a name="l00295"></a><span class="lineno">  295</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(out)</span>,   <span class="keywordtype">optional</span> :: arr_max_levels_out(nflds)</div><div class="line"><a name="l00296"></a><span class="lineno">  296</span>&#160;                                         <span class="comment">!&lt; output of number of levels of</span></div><div class="line"><a name="l00297"></a><span class="lineno">  297</span>&#160;<span class="comment">                                         !&lt; integer expansion to used</span></div><div class="line"><a name="l00298"></a><span class="lineno">  298</span>&#160;</div><div class="line"><a name="l00299"></a><span class="lineno">  299</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: gbl_max_nsummands</div><div class="line"><a name="l00300"></a><span class="lineno">  300</span>&#160;                                         <span class="comment">!&lt; maximum of nsummand over all</span></div><div class="line"><a name="l00301"></a><span class="lineno">  301</span>&#160;<span class="comment">                                         !&lt; processes</span></div><div class="line"><a name="l00302"></a><span class="lineno">  302</span>&#160;</div><div class="line"><a name="l00303"></a><span class="lineno">  303</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(out)</span>,   <span class="keywordtype">optional</span> :: gbl_max_nsummands_out</div><div class="line"><a name="l00304"></a><span class="lineno">  304</span>&#160;                                         <span class="comment">!&lt; calculated maximum nsummands</span></div><div class="line"><a name="l00305"></a><span class="lineno">  305</span>&#160;<span class="comment">                                         !&lt; over all processes</span></div><div class="line"><a name="l00306"></a><span class="lineno">  306</span>&#160;</div><div class="line"><a name="l00307"></a><span class="lineno">  307</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: gbl_count  </div><div class="line"><a name="l00308"></a><span class="lineno">  308</span>&#160;                                         <span class="comment">!&lt; was total number of summands;</span></div><div class="line"><a name="l00309"></a><span class="lineno">  309</span>&#160;<span class="comment">                                         !&lt; now is ignored; use </span></div><div class="line"><a name="l00310"></a><span class="lineno">  310</span>&#160;<span class="comment">                                         !&lt; gbl_max_nsummands instead</span></div><div class="line"><a name="l00311"></a><span class="lineno">  311</span>&#160;</div><div class="line"><a name="l00312"></a><span class="lineno">  312</span>&#160;      <span class="keywordtype">logical</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: repro_sum_validate</div><div class="line"><a name="l00313"></a><span class="lineno">  313</span>&#160;         <span class="comment">!&lt; flag enabling/disabling testing that gmax and  max_levels are </span></div><div class="line"><a name="l00314"></a><span class="lineno">  314</span>&#160;<span class="comment">         !&lt; accurate/sufficient. Default is enabled.</span></div><div class="line"><a name="l00315"></a><span class="lineno">  315</span>&#160;</div><div class="line"><a name="l00316"></a><span class="lineno">  316</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(inout)</span>, <span class="keywordtype">optional</span> :: repro_sum_stats(5)</div><div class="line"><a name="l00317"></a><span class="lineno">  317</span>&#160;                                   <span class="comment">!&lt; increment running totals for</span></div><div class="line"><a name="l00318"></a><span class="lineno">  318</span>&#160;<span class="comment">                                   !&lt;  (1) one-reduction repro_sum</span></div><div class="line"><a name="l00319"></a><span class="lineno">  319</span>&#160;<span class="comment">                                   !&lt;  (2) two-reduction repro_sum</span></div><div class="line"><a name="l00320"></a><span class="lineno">  320</span>&#160;<span class="comment">                                   !&lt;  (3) both types in one call</span></div><div class="line"><a name="l00321"></a><span class="lineno">  321</span>&#160;<span class="comment">                                   !&lt;  (4) nonrepro_sum</span></div><div class="line"><a name="l00322"></a><span class="lineno">  322</span>&#160;<span class="comment">                                   !&lt;  (5) global max nsummands reduction</span></div><div class="line"><a name="l00323"></a><span class="lineno">  323</span>&#160;</div><div class="line"><a name="l00324"></a><span class="lineno">  324</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>,   <span class="keywordtype">optional</span> :: rel_diff(2,nflds)</div><div class="line"><a name="l00325"></a><span class="lineno">  325</span>&#160;                                         <span class="comment">!&lt; relative and absolute</span></div><div class="line"><a name="l00326"></a><span class="lineno">  326</span>&#160;<span class="comment">                                         !&lt;  differences between fixed </span></div><div class="line"><a name="l00327"></a><span class="lineno">  327</span>&#160;<span class="comment">                                         !&lt;  and floating point sums</span></div><div class="line"><a name="l00328"></a><span class="lineno">  328</span>&#160;</div><div class="line"><a name="l00329"></a><span class="lineno">  329</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span>,    <span class="keywordtype">optional</span> :: commid     </div><div class="line"><a name="l00330"></a><span class="lineno">  330</span>&#160;                                         <span class="comment">!&lt; MPI communicator</span></div><div class="line"><a name="l00331"></a><span class="lineno">  331</span>&#160;</div><div class="line"><a name="l00332"></a><span class="lineno">  332</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l00333"></a><span class="lineno">  333</span>&#160;</div><div class="line"><a name="l00334"></a><span class="lineno">  334</span>&#160;      <span class="keywordtype">logical</span> :: use_ddpdd_sum           <span class="comment">! flag indicating whether to</span></div><div class="line"><a name="l00335"></a><span class="lineno">  335</span>&#160;                                         <span class="comment">!  use oasis_reprosum_ddpdd or not</span></div><div class="line"><a name="l00336"></a><span class="lineno">  336</span>&#160;      <span class="keywordtype">logical</span> :: recompute               <span class="comment">! flag indicating need to </span></div><div class="line"><a name="l00337"></a><span class="lineno">  337</span>&#160;                                         <span class="comment">!  determine gmax/gmin before </span></div><div class="line"><a name="l00338"></a><span class="lineno">  338</span>&#160;                                         <span class="comment">!  computing sum</span></div><div class="line"><a name="l00339"></a><span class="lineno">  339</span>&#160;      <span class="keywordtype">logical</span> :: validate                <span class="comment">! flag indicating need to </span></div><div class="line"><a name="l00340"></a><span class="lineno">  340</span>&#160;                                         <span class="comment">!  verify gmax and max_levels </span></div><div class="line"><a name="l00341"></a><span class="lineno">  341</span>&#160;                                         <span class="comment">!  are accurate/sufficient</span></div><div class="line"><a name="l00342"></a><span class="lineno">  342</span>&#160;      <span class="keywordtype">integer</span> :: omp_nthreads            <span class="comment">! number of OpenMP threads</span></div><div class="line"><a name="l00343"></a><span class="lineno">  343</span>&#160;      <span class="keywordtype">integer</span> :: mpi_comm                <span class="comment">! MPI subcommunicator</span></div><div class="line"><a name="l00344"></a><span class="lineno">  344</span>&#160;      <span class="keywordtype">integer</span> :: tasks                   <span class="comment">! number of MPI processes</span></div><div class="line"><a name="l00345"></a><span class="lineno">  345</span>&#160;      <span class="keywordtype">integer</span> :: ierr                    <span class="comment">! MPI error return</span></div><div class="line"><a name="l00346"></a><span class="lineno">  346</span>&#160;      <span class="keywordtype">integer</span> :: ifld, isum, ithread     <span class="comment">! loop variables</span></div><div class="line"><a name="l00347"></a><span class="lineno">  347</span>&#160;      <span class="keywordtype">integer</span> :: max_nsummands           <span class="comment">! max nsummands over all processes</span></div><div class="line"><a name="l00348"></a><span class="lineno">  348</span>&#160;                                         <span class="comment">!  or threads (used in both ways)</span></div><div class="line"><a name="l00349"></a><span class="lineno">  349</span>&#160;</div><div class="line"><a name="l00350"></a><span class="lineno">  350</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">allocatable</span> :: isum_beg(:), isum_end(:)</div><div class="line"><a name="l00351"></a><span class="lineno">  351</span>&#160;                                         <span class="comment">! range of summand indices for each</span></div><div class="line"><a name="l00352"></a><span class="lineno">  352</span>&#160;                                         <span class="comment">!  OpenMP thread</span></div><div class="line"><a name="l00353"></a><span class="lineno">  353</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">allocatable</span> :: arr_tlmin_exp(:,:)</div><div class="line"><a name="l00354"></a><span class="lineno">  354</span>&#160;                                         <span class="comment">! per thread local exponent minima</span></div><div class="line"><a name="l00355"></a><span class="lineno">  355</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">allocatable</span> :: arr_tlmax_exp(:,:)</div><div class="line"><a name="l00356"></a><span class="lineno">  356</span>&#160;                                         <span class="comment">! per thread local exponent maxima</span></div><div class="line"><a name="l00357"></a><span class="lineno">  357</span>&#160;      <span class="keywordtype">integer</span> :: arr_exp, arr_exp_tlmin, arr_exp_tlmax</div><div class="line"><a name="l00358"></a><span class="lineno">  358</span>&#160;                                         <span class="comment">! summand exponent and working min/max</span></div><div class="line"><a name="l00359"></a><span class="lineno">  359</span>&#160;      <span class="keywordtype">integer</span> :: arr_lmin_exp(nflds)     <span class="comment">! local exponent minima</span></div><div class="line"><a name="l00360"></a><span class="lineno">  360</span>&#160;      <span class="keywordtype">integer</span> :: arr_lmax_exp(nflds)     <span class="comment">! local exponent maxima</span></div><div class="line"><a name="l00361"></a><span class="lineno">  361</span>&#160;      <span class="keywordtype">integer</span> :: arr_lextremes(0:nflds,2)<span class="comment">! local exponent extrema</span></div><div class="line"><a name="l00362"></a><span class="lineno">  362</span>&#160;      <span class="keywordtype">integer</span> :: arr_gextremes(0:nflds,2)<span class="comment">! global exponent extrema</span></div><div class="line"><a name="l00363"></a><span class="lineno">  363</span>&#160;</div><div class="line"><a name="l00364"></a><span class="lineno">  364</span>&#160;      <span class="keywordtype">integer</span> :: arr_gmax_exp(nflds)     <span class="comment">! global exponents maxima </span></div><div class="line"><a name="l00365"></a><span class="lineno">  365</span>&#160;      <span class="keywordtype">integer</span> :: arr_gmin_exp(nflds)     <span class="comment">! global exponents minima</span></div><div class="line"><a name="l00366"></a><span class="lineno">  366</span>&#160;      <span class="keywordtype">integer</span> :: arr_max_shift           <span class="comment">! maximum safe exponent for </span></div><div class="line"><a name="l00367"></a><span class="lineno">  367</span>&#160;                                         <span class="comment">!  value &lt; 1 (so that sum does</span></div><div class="line"><a name="l00368"></a><span class="lineno">  368</span>&#160;                                         <span class="comment">!  not overflow)</span></div><div class="line"><a name="l00369"></a><span class="lineno">  369</span>&#160;      <span class="keywordtype">integer</span> :: max_levels(nflds)       <span class="comment">! maximum number of levels of </span></div><div class="line"><a name="l00370"></a><span class="lineno">  370</span>&#160;                                         <span class="comment">!  integer expansion to use</span></div><div class="line"><a name="l00371"></a><span class="lineno">  371</span>&#160;      <span class="keywordtype">integer</span> :: max_level               <span class="comment">! maximum value in max_levels</span></div><div class="line"><a name="l00372"></a><span class="lineno">  372</span>&#160;      <span class="keywordtype">integer</span> :: gbl_max_red             <span class="comment">! global max local sum reduction? (0/1)</span></div><div class="line"><a name="l00373"></a><span class="lineno">  373</span>&#160;      <span class="keywordtype">integer</span> :: repro_sum_fast          <span class="comment">! 1 reduction repro_sum? (0/1)</span></div><div class="line"><a name="l00374"></a><span class="lineno">  374</span>&#160;      <span class="keywordtype">integer</span> :: repro_sum_slow          <span class="comment">! 2 reduction repro_sum? (0/1)</span></div><div class="line"><a name="l00375"></a><span class="lineno">  375</span>&#160;      <span class="keywordtype">integer</span> :: repro_sum_both          <span class="comment">! both fast and slow? (0/1)</span></div><div class="line"><a name="l00376"></a><span class="lineno">  376</span>&#160;      <span class="keywordtype">integer</span> :: nonrepro_sum            <span class="comment">! nonrepro_sum? (0/1)</span></div><div class="line"><a name="l00377"></a><span class="lineno">  377</span>&#160;</div><div class="line"><a name="l00378"></a><span class="lineno">  378</span>&#160;      <span class="keywordtype">real(r8)</span> :: xmax_nsummands         <span class="comment">! dble of max_nsummands</span></div><div class="line"><a name="l00379"></a><span class="lineno">  379</span>&#160;      <span class="keywordtype">real(r8)</span> :: arr_lsum(nflds)        <span class="comment">! local sums</span></div><div class="line"><a name="l00380"></a><span class="lineno">  380</span>&#160;      <span class="keywordtype">real(r8)</span> :: arr_gsum_fast(nflds)   <span class="comment">! global sum calculated using </span></div><div class="line"><a name="l00381"></a><span class="lineno">  381</span>&#160;                                         <span class="comment">!  fast, nonreproducible, </span></div><div class="line"><a name="l00382"></a><span class="lineno">  382</span>&#160;                                         <span class="comment">!  floating point alg.</span></div><div class="line"><a name="l00383"></a><span class="lineno">  383</span>&#160;      <span class="keywordtype">real(r8)</span> :: abs_diff               <span class="comment">! absolute difference between </span></div><div class="line"><a name="l00384"></a><span class="lineno">  384</span>&#160;                                         <span class="comment">!  fixed and floating point </span></div><div class="line"><a name="l00385"></a><span class="lineno">  385</span>&#160;                                         <span class="comment">!  sums</span></div><div class="line"><a name="l00386"></a><span class="lineno">  386</span>&#160;<span class="preprocessor">#ifdef _OPENMP</span></div><div class="line"><a name="l00387"></a><span class="lineno">  387</span>&#160;<span class="preprocessor"></span>      <span class="keywordtype">integer</span> omp_get_max_threads</div><div class="line"><a name="l00388"></a><span class="lineno">  388</span>&#160;      <span class="keywordtype">external</span> omp_get_max_threads</div><div class="line"><a name="l00389"></a><span class="lineno">  389</span>&#160;<span class="preprocessor">#endif  </span></div><div class="line"><a name="l00390"></a><span class="lineno">  390</span>&#160;<span class="preprocessor"></span>      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_calc)&#39;</span></div><div class="line"><a name="l00391"></a><span class="lineno">  391</span>&#160;</div><div class="line"><a name="l00392"></a><span class="lineno">  392</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00393"></a><span class="lineno">  393</span>&#160;</div><div class="line"><a name="l00394"></a><span class="lineno">  394</span>&#160;<span class="comment">! check whether should use oasis_reprosum_ddpdd algorithm</span></div><div class="line"><a name="l00395"></a><span class="lineno">  395</span>&#160;      use_ddpdd_sum = <a class="code" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">repro_sum_use_ddpdd</a></div><div class="line"><a name="l00396"></a><span class="lineno">  396</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(ddpdd_sum) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00397"></a><span class="lineno">  397</span>&#160;         use_ddpdd_sum = ddpdd_sum</div><div class="line"><a name="l00398"></a><span class="lineno">  398</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00399"></a><span class="lineno">  399</span>&#160;</div><div class="line"><a name="l00400"></a><span class="lineno">  400</span>&#160;<span class="comment">! check whether intrinsic-based algorithm will work on this system</span></div><div class="line"><a name="l00401"></a><span class="lineno">  401</span>&#160;<span class="comment">! (requires floating point and integer bases to be the same)</span></div><div class="line"><a name="l00402"></a><span class="lineno">  402</span>&#160;<span class="comment">! If not, always use ddpdd.</span></div><div class="line"><a name="l00403"></a><span class="lineno">  403</span>&#160;      use_ddpdd_sum = use_ddpdd_sum .or. (radix(0._r8) /= radix(0_i8))</div><div class="line"><a name="l00404"></a><span class="lineno">  404</span>&#160;</div><div class="line"><a name="l00405"></a><span class="lineno">  405</span>&#160;<span class="comment">! initialize local statistics variables</span></div><div class="line"><a name="l00406"></a><span class="lineno">  406</span>&#160;      gbl_max_red = 0</div><div class="line"><a name="l00407"></a><span class="lineno">  407</span>&#160;      repro_sum_fast = 0</div><div class="line"><a name="l00408"></a><span class="lineno">  408</span>&#160;      repro_sum_slow = 0</div><div class="line"><a name="l00409"></a><span class="lineno">  409</span>&#160;      repro_sum_both = 0</div><div class="line"><a name="l00410"></a><span class="lineno">  410</span>&#160;      nonrepro_sum = 0</div><div class="line"><a name="l00411"></a><span class="lineno">  411</span>&#160;</div><div class="line"><a name="l00412"></a><span class="lineno">  412</span>&#160;<span class="comment">! set MPI communicator</span></div><div class="line"><a name="l00413"></a><span class="lineno">  413</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(commid) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00414"></a><span class="lineno">  414</span>&#160;         mpi_comm = commid</div><div class="line"><a name="l00415"></a><span class="lineno">  415</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l00416"></a><span class="lineno">  416</span>&#160;         mpi_comm = mpi_comm_world</div><div class="line"><a name="l00417"></a><span class="lineno">  417</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00418"></a><span class="lineno">  418</span>&#160;</div><div class="line"><a name="l00419"></a><span class="lineno">  419</span>&#160;      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keywordflow">then</span></div><div class="line"><a name="l00420"></a><span class="lineno">  420</span>&#160;         <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;oasis_reprosum_prebarrier&#39;</span>)</div><div class="line"><a name="l00421"></a><span class="lineno">  421</span>&#160;         <span class="keyword">call </span><a class="code" href="namespacemod__oasis__mpi.html#aaf477847380fbbbc1341520b0b3e66da">oasis_mpi_barrier</a>(mpi_comm,subname)</div><div class="line"><a name="l00422"></a><span class="lineno">  422</span>&#160;         <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a> (<span class="stringliteral">&#39;oasis_reprosum_prebarrier&#39;</span>)</div><div class="line"><a name="l00423"></a><span class="lineno">  423</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00424"></a><span class="lineno">  424</span>&#160;</div><div class="line"><a name="l00425"></a><span class="lineno">  425</span>&#160;      <span class="keywordflow">if</span> ( use_ddpdd_sum ) <span class="keywordflow">then</span></div><div class="line"><a name="l00426"></a><span class="lineno">  426</span>&#160;</div><div class="line"><a name="l00427"></a><span class="lineno">  427</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;oasis_reprosum_ddpdd&#39;</span>)</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="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">oasis_reprosum_ddpdd</a>(arr, arr_gsum, nsummands, dsummands, &amp;</div><div class="line"><a name="l00430"></a><span class="lineno">  430</span>&#160;                              nflds, mpi_comm)</div><div class="line"><a name="l00431"></a><span class="lineno">  431</span>&#160;         repro_sum_fast = 1</div><div class="line"><a name="l00432"></a><span class="lineno">  432</span>&#160;</div><div class="line"><a name="l00433"></a><span class="lineno">  433</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&#39;oasis_reprosum_ddpdd&#39;</span>)</div><div class="line"><a name="l00434"></a><span class="lineno">  434</span>&#160;</div><div class="line"><a name="l00435"></a><span class="lineno">  435</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l00436"></a><span class="lineno">  436</span>&#160;</div><div class="line"><a name="l00437"></a><span class="lineno">  437</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;oasis_reprosum_int&#39;</span>)</div><div class="line"><a name="l00438"></a><span class="lineno">  438</span>&#160;</div><div class="line"><a name="l00439"></a><span class="lineno">  439</span>&#160;<span class="comment">! get number of MPI tasks</span></div><div class="line"><a name="l00440"></a><span class="lineno">  440</span>&#160;         <span class="keyword">call </span>mpi_comm_size(mpi_comm, tasks, ierr)</div><div class="line"><a name="l00441"></a><span class="lineno">  441</span>&#160;</div><div class="line"><a name="l00442"></a><span class="lineno">  442</span>&#160;<span class="comment">! get number of OpenMP threads</span></div><div class="line"><a name="l00443"></a><span class="lineno">  443</span>&#160;<span class="preprocessor">#ifdef _OPENMP</span></div><div class="line"><a name="l00444"></a><span class="lineno">  444</span>&#160;<span class="preprocessor"></span>         omp_nthreads = omp_get_max_threads()</div><div class="line"><a name="l00445"></a><span class="lineno">  445</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l00446"></a><span class="lineno">  446</span>&#160;<span class="preprocessor"></span>         omp_nthreads = 1</div><div class="line"><a name="l00447"></a><span class="lineno">  447</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l00448"></a><span class="lineno">  448</span>&#160;<span class="preprocessor"></span></div><div class="line"><a name="l00449"></a><span class="lineno">  449</span>&#160;<span class="comment">! see if have sufficient information to not require max/min allreduce</span></div><div class="line"><a name="l00450"></a><span class="lineno">  450</span>&#160;         recompute = .true.</div><div class="line"><a name="l00451"></a><span class="lineno">  451</span>&#160;         validate = .false.</div><div class="line"><a name="l00452"></a><span class="lineno">  452</span>&#160;         <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_gbl_max) .and. <span class="keyword">present</span>(arr_max_levels) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00453"></a><span class="lineno">  453</span>&#160;            recompute = .false.</div><div class="line"><a name="l00454"></a><span class="lineno">  454</span>&#160;</div><div class="line"><a name="l00455"></a><span class="lineno">  455</span>&#160;<span class="comment">! setting lower bound on max_level*nflds to be 64 to improve OpenMP</span></div><div class="line"><a name="l00456"></a><span class="lineno">  456</span>&#160;<span class="comment">! performance for loopb in oasis_reprosum_int</span></div><div class="line"><a name="l00457"></a><span class="lineno">  457</span>&#160;            max_level = (64/nflds) + 1</div><div class="line"><a name="l00458"></a><span class="lineno">  458</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00459"></a><span class="lineno">  459</span>&#160;               <span class="keywordflow">if</span> ((arr_gbl_max(ifld) .ge. 0.0_r8) .and. &amp;</div><div class="line"><a name="l00460"></a><span class="lineno">  460</span>&#160;                   (arr_max_levels(ifld) &gt; 0)) <span class="keywordflow">then</span></div><div class="line"><a name="l00461"></a><span class="lineno">  461</span>&#160;</div><div class="line"><a name="l00462"></a><span class="lineno">  462</span>&#160;                  arr_gmax_exp(ifld)  = exponent(arr_gbl_max(ifld))</div><div class="line"><a name="l00463"></a><span class="lineno">  463</span>&#160;                  <span class="keywordflow">if</span> (max_level &lt; arr_max_levels(ifld)) &amp;</div><div class="line"><a name="l00464"></a><span class="lineno">  464</span>&#160;                     max_level = arr_max_levels(ifld)</div><div class="line"><a name="l00465"></a><span class="lineno">  465</span>&#160;</div><div class="line"><a name="l00466"></a><span class="lineno">  466</span>&#160;               <span class="keywordflow">else</span></div><div class="line"><a name="l00467"></a><span class="lineno">  467</span>&#160;                  recompute = .true.</div><div class="line"><a name="l00468"></a><span class="lineno">  468</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00469"></a><span class="lineno">  469</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l00470"></a><span class="lineno">  470</span>&#160;</div><div class="line"><a name="l00471"></a><span class="lineno">  471</span>&#160;            <span class="keywordflow">if</span> (.not. recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00472"></a><span class="lineno">  472</span>&#160;</div><div class="line"><a name="l00473"></a><span class="lineno">  473</span>&#160;<span class="comment">! determine maximum number of summands in local phases of the </span></div><div class="line"><a name="l00474"></a><span class="lineno">  474</span>&#160;<span class="comment">! algorithm</span></div><div class="line"><a name="l00475"></a><span class="lineno">  475</span>&#160;               <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&quot;repro_sum_allr_max&quot;</span>)</div><div class="line"><a name="l00476"></a><span class="lineno">  476</span>&#160;               <span class="keywordflow">if</span> ( <span class="keyword">present</span>(gbl_max_nsummands) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00477"></a><span class="lineno">  477</span>&#160;                  <span class="keywordflow">if</span> (gbl_max_nsummands &lt; 1) <span class="keywordflow">then</span></div><div class="line"><a name="l00478"></a><span class="lineno">  478</span>&#160;                     <span class="keyword">call </span>mpi_allreduce (nsummands, max_nsummands, 1, &amp;</div><div class="line"><a name="l00479"></a><span class="lineno">  479</span>&#160;                                         mpi_integer, mpi_max, mpi_comm, ierr)</div><div class="line"><a name="l00480"></a><span class="lineno">  480</span>&#160;                     gbl_max_red = 1</div><div class="line"><a name="l00481"></a><span class="lineno">  481</span>&#160;                  <span class="keywordflow">else</span></div><div class="line"><a name="l00482"></a><span class="lineno">  482</span>&#160;                     max_nsummands = gbl_max_nsummands</div><div class="line"><a name="l00483"></a><span class="lineno">  483</span>&#160;<span class="keywordflow">                  endif</span></div><div class="line"><a name="l00484"></a><span class="lineno">  484</span>&#160;               <span class="keywordflow">else</span></div><div class="line"><a name="l00485"></a><span class="lineno">  485</span>&#160;                  <span class="keyword">call </span>mpi_allreduce (nsummands, max_nsummands, 1, &amp;</div><div class="line"><a name="l00486"></a><span class="lineno">  486</span>&#160;                                      mpi_integer, mpi_max, mpi_comm, ierr)</div><div class="line"><a name="l00487"></a><span class="lineno">  487</span>&#160;                  gbl_max_red = 1</div><div class="line"><a name="l00488"></a><span class="lineno">  488</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00489"></a><span class="lineno">  489</span>&#160;               <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&quot;repro_sum_allr_max&quot;</span>)</div><div class="line"><a name="l00490"></a><span class="lineno">  490</span>&#160;</div><div class="line"><a name="l00491"></a><span class="lineno">  491</span>&#160;<span class="comment">! determine maximum shift. Shift needs to be small enough that summation </span></div><div class="line"><a name="l00492"></a><span class="lineno">  492</span>&#160;<span class="comment">!  does not exceed maximum number of digits in i8.</span></div><div class="line"><a name="l00493"></a><span class="lineno">  493</span>&#160;</div><div class="line"><a name="l00494"></a><span class="lineno">  494</span>&#160;<span class="comment">! if requested, return max_nsummands before it is redefined</span></div><div class="line"><a name="l00495"></a><span class="lineno">  495</span>&#160;               <span class="keywordflow">if</span> ( <span class="keyword">present</span>( gbl_max_nsummands_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00496"></a><span class="lineno">  496</span>&#160;                  gbl_max_nsummands_out = max_nsummands</div><div class="line"><a name="l00497"></a><span class="lineno">  497</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00498"></a><span class="lineno">  498</span>&#160;</div><div class="line"><a name="l00499"></a><span class="lineno">  499</span>&#160;<span class="comment">! summing within each thread first</span></div><div class="line"><a name="l00500"></a><span class="lineno">  500</span>&#160;               max_nsummands = (max_nsummands/omp_nthreads) + 1</div><div class="line"><a name="l00501"></a><span class="lineno">  501</span>&#160;<span class="comment">! then over threads and tasks</span></div><div class="line"><a name="l00502"></a><span class="lineno">  502</span>&#160;               max_nsummands = max(max_nsummands, tasks*omp_nthreads)</div><div class="line"><a name="l00503"></a><span class="lineno">  503</span>&#160;</div><div class="line"><a name="l00504"></a><span class="lineno">  504</span>&#160;               xmax_nsummands = dble(max_nsummands)</div><div class="line"><a name="l00505"></a><span class="lineno">  505</span>&#160;               arr_max_shift = digits(0_i8) - (exponent(xmax_nsummands) + 1)</div><div class="line"><a name="l00506"></a><span class="lineno">  506</span>&#160;               <span class="keywordflow">if</span> (arr_max_shift &lt; 2) <span class="keywordflow">then</span></div><div class="line"><a name="l00507"></a><span class="lineno">  507</span>&#160;                  <span class="keyword">write</span>(<a class="code" href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">nulprt</a>,*) subname,<a class="code" href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">estr</a>,<span class="stringliteral">&#39; number of summands too large for fixed precision algorithm&#39;</span></div><div class="line"><a name="l00508"></a><span class="lineno">  508</span>&#160;                  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__sys.html#af1943a0ce0c5765ba8a714379065464f">oasis_abort</a>(file=__file__,line=__line__)</div><div class="line"><a name="l00509"></a><span class="lineno">  509</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00510"></a><span class="lineno">  510</span>&#160;</div><div class="line"><a name="l00511"></a><span class="lineno">  511</span>&#160;<span class="comment">! calculate sum</span></div><div class="line"><a name="l00512"></a><span class="lineno">  512</span>&#160;               <span class="keywordflow">if</span> (<span class="keyword">present</span>(repro_sum_validate)) <span class="keywordflow">then</span></div><div class="line"><a name="l00513"></a><span class="lineno">  513</span>&#160;                  validate = repro_sum_validate</div><div class="line"><a name="l00514"></a><span class="lineno">  514</span>&#160;               <span class="keywordflow">else</span></div><div class="line"><a name="l00515"></a><span class="lineno">  515</span>&#160;                  validate = .true.</div><div class="line"><a name="l00516"></a><span class="lineno">  516</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00517"></a><span class="lineno">  517</span>&#160;               <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a>(arr, arr_gsum, nsummands, dsummands, &amp;</div><div class="line"><a name="l00518"></a><span class="lineno">  518</span>&#160;                                     nflds, arr_max_shift, arr_gmax_exp, &amp;</div><div class="line"><a name="l00519"></a><span class="lineno">  519</span>&#160;                                     arr_max_levels, max_level, validate, &amp;</div><div class="line"><a name="l00520"></a><span class="lineno">  520</span>&#160;                                     recompute, omp_nthreads, mpi_comm)</div><div class="line"><a name="l00521"></a><span class="lineno">  521</span>&#160;</div><div class="line"><a name="l00522"></a><span class="lineno">  522</span>&#160;<span class="comment">! record statistics, etc.</span></div><div class="line"><a name="l00523"></a><span class="lineno">  523</span>&#160;               repro_sum_fast = 1</div><div class="line"><a name="l00524"></a><span class="lineno">  524</span>&#160;               <span class="keywordflow">if</span> (recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00525"></a><span class="lineno">  525</span>&#160;                  repro_sum_both = 1</div><div class="line"><a name="l00526"></a><span class="lineno">  526</span>&#160;               <span class="keywordflow">else</span> </div><div class="line"><a name="l00527"></a><span class="lineno">  527</span>&#160;<span class="comment">! if requested, return specified levels and upper bounds on maxima</span></div><div class="line"><a name="l00528"></a><span class="lineno">  528</span>&#160;                  <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_max_levels_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00529"></a><span class="lineno">  529</span>&#160;                     <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00530"></a><span class="lineno">  530</span>&#160;                        arr_max_levels_out(ifld) = arr_max_levels(ifld)</div><div class="line"><a name="l00531"></a><span class="lineno">  531</span>&#160;<span class="keywordflow">                     enddo</span></div><div class="line"><a name="l00532"></a><span class="lineno">  532</span>&#160;<span class="keywordflow">                  endif</span></div><div class="line"><a name="l00533"></a><span class="lineno">  533</span>&#160;                  <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_gbl_max_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00534"></a><span class="lineno">  534</span>&#160;                     <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00535"></a><span class="lineno">  535</span>&#160;                       arr_gbl_max_out(ifld) = arr_gbl_max(ifld)</div><div class="line"><a name="l00536"></a><span class="lineno">  536</span>&#160;<span class="keywordflow">                     enddo</span></div><div class="line"><a name="l00537"></a><span class="lineno">  537</span>&#160;<span class="keywordflow">                  endif</span></div><div class="line"><a name="l00538"></a><span class="lineno">  538</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00539"></a><span class="lineno">  539</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00540"></a><span class="lineno">  540</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l00541"></a><span class="lineno">  541</span>&#160;</div><div class="line"><a name="l00542"></a><span class="lineno">  542</span>&#160;<span class="comment">! do not have sufficient information; calculate global max/min and </span></div><div class="line"><a name="l00543"></a><span class="lineno">  543</span>&#160;<span class="comment">! use to compute required number of levels</span></div><div class="line"><a name="l00544"></a><span class="lineno">  544</span>&#160;         <span class="keywordflow">if</span> (recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00545"></a><span class="lineno">  545</span>&#160;</div><div class="line"><a name="l00546"></a><span class="lineno">  546</span>&#160;<span class="comment">! record statistic</span></div><div class="line"><a name="l00547"></a><span class="lineno">  547</span>&#160;            repro_sum_slow = 1</div><div class="line"><a name="l00548"></a><span class="lineno">  548</span>&#160;</div><div class="line"><a name="l00549"></a><span class="lineno">  549</span>&#160;<span class="comment">! determine maximum and minimum (non-zero) summand values and</span></div><div class="line"><a name="l00550"></a><span class="lineno">  550</span>&#160;<span class="comment">! maximum number of local summands</span></div><div class="line"><a name="l00551"></a><span class="lineno">  551</span>&#160;</div><div class="line"><a name="l00552"></a><span class="lineno">  552</span>&#160;<span class="comment">! allocate thread-specific work space</span></div><div class="line"><a name="l00553"></a><span class="lineno">  553</span>&#160;            <span class="keyword">allocate</span>(arr_tlmax_exp(nflds,omp_nthreads))</div><div class="line"><a name="l00554"></a><span class="lineno">  554</span>&#160;            <span class="keyword">allocate</span>(arr_tlmin_exp(nflds,omp_nthreads))</div><div class="line"><a name="l00555"></a><span class="lineno">  555</span>&#160;            <span class="keyword">allocate</span>(isum_beg(omp_nthreads))</div><div class="line"><a name="l00556"></a><span class="lineno">  556</span>&#160;            <span class="keyword">allocate</span>(isum_end(omp_nthreads))</div><div class="line"><a name="l00557"></a><span class="lineno">  557</span>&#160;</div><div class="line"><a name="l00558"></a><span class="lineno">  558</span>&#160;<span class="comment">! split summand index range over OpenMP threads</span></div><div class="line"><a name="l00559"></a><span class="lineno">  559</span>&#160;            <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">split_indices</a>(nsummands, omp_nthreads, isum_beg, isum_end)</div><div class="line"><a name="l00560"></a><span class="lineno">  560</span>&#160;</div><div class="line"><a name="l00561"></a><span class="lineno">  561</span>&#160;<span class="comment">!$omp parallel do      &amp;</span></div><div class="line"><a name="l00562"></a><span class="lineno">  562</span>&#160;<span class="comment">!$omp default(shared)  &amp;</span></div><div class="line"><a name="l00563"></a><span class="lineno">  563</span>&#160;<span class="comment">!$omp private(ithread, ifld, isum, arr_exp, arr_exp_tlmin, arr_exp_tlmax)</span></div><div class="line"><a name="l00564"></a><span class="lineno">  564</span>&#160;            <span class="keywordflow">do</span> ithread=1,omp_nthreads</div><div class="line"><a name="l00565"></a><span class="lineno">  565</span>&#160;               <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;repro_sum_loopa&#39;</span>)</div><div class="line"><a name="l00566"></a><span class="lineno">  566</span>&#160;               <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00567"></a><span class="lineno">  567</span>&#160;                  arr_exp_tlmin = maxexponent(1._r8)</div><div class="line"><a name="l00568"></a><span class="lineno">  568</span>&#160;                  arr_exp_tlmax = minexponent(1._r8)</div><div class="line"><a name="l00569"></a><span class="lineno">  569</span>&#160;                  <span class="keywordflow">do</span> isum=isum_beg(ithread),isum_end(ithread)</div><div class="line"><a name="l00570"></a><span class="lineno">  570</span>&#160;                     <span class="keywordflow">if</span> (arr(isum,ifld) .ne. 0.0_r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00571"></a><span class="lineno">  571</span>&#160;                        arr_exp = exponent(arr(isum,ifld))</div><div class="line"><a name="l00572"></a><span class="lineno">  572</span>&#160;                        arr_exp_tlmin = min(arr_exp,arr_exp_tlmin)</div><div class="line"><a name="l00573"></a><span class="lineno">  573</span>&#160;                        arr_exp_tlmax = max(arr_exp,arr_exp_tlmax)</div><div class="line"><a name="l00574"></a><span class="lineno">  574</span>&#160;<span class="keywordflow">                     endif</span></div><div class="line"><a name="l00575"></a><span class="lineno">  575</span>&#160;<span class="keywordflow">                  end do</span></div><div class="line"><a name="l00576"></a><span class="lineno">  576</span>&#160;                  arr_tlmin_exp(ifld,ithread) = arr_exp_tlmin</div><div class="line"><a name="l00577"></a><span class="lineno">  577</span>&#160;                  arr_tlmax_exp(ifld,ithread) = arr_exp_tlmax</div><div class="line"><a name="l00578"></a><span class="lineno">  578</span>&#160;<span class="keywordflow">               end do</span></div><div class="line"><a name="l00579"></a><span class="lineno">  579</span>&#160;               <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&#39;repro_sum_loopa&#39;</span>)</div><div class="line"><a name="l00580"></a><span class="lineno">  580</span>&#160;<span class="keywordflow">            end do</span></div><div class="line"><a name="l00581"></a><span class="lineno">  581</span>&#160;</div><div class="line"><a name="l00582"></a><span class="lineno">  582</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00583"></a><span class="lineno">  583</span>&#160;               arr_lmax_exp(ifld) = maxval(arr_tlmax_exp(ifld,:))</div><div class="line"><a name="l00584"></a><span class="lineno">  584</span>&#160;               arr_lmin_exp(ifld) = minval(arr_tlmin_exp(ifld,:))</div><div class="line"><a name="l00585"></a><span class="lineno">  585</span>&#160;<span class="keywordflow">            end do</span></div><div class="line"><a name="l00586"></a><span class="lineno">  586</span>&#160;            <span class="keyword">deallocate</span>(arr_tlmin_exp,arr_tlmax_exp,isum_beg,isum_end)</div><div class="line"><a name="l00587"></a><span class="lineno">  587</span>&#160;</div><div class="line"><a name="l00588"></a><span class="lineno">  588</span>&#160;            arr_lextremes(0,:) = -nsummands</div><div class="line"><a name="l00589"></a><span class="lineno">  589</span>&#160;            arr_lextremes(1:nflds,1) = -arr_lmax_exp(:)</div><div class="line"><a name="l00590"></a><span class="lineno">  590</span>&#160;            arr_lextremes(1:nflds,2) = arr_lmin_exp(:)</div><div class="line"><a name="l00591"></a><span class="lineno">  591</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&quot;repro_sum_allr_minmax&quot;</span>)</div><div class="line"><a name="l00592"></a><span class="lineno">  592</span>&#160;            <span class="keyword">call </span>mpi_allreduce (arr_lextremes, arr_gextremes, 2*(nflds+1), &amp;</div><div class="line"><a name="l00593"></a><span class="lineno">  593</span>&#160;                                mpi_integer, mpi_min, mpi_comm, ierr)</div><div class="line"><a name="l00594"></a><span class="lineno">  594</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&quot;repro_sum_allr_minmax&quot;</span>)</div><div class="line"><a name="l00595"></a><span class="lineno">  595</span>&#160;            max_nsummands   = -arr_gextremes(0,1)</div><div class="line"><a name="l00596"></a><span class="lineno">  596</span>&#160;            arr_gmax_exp(:) = -arr_gextremes(1:nflds,1)</div><div class="line"><a name="l00597"></a><span class="lineno">  597</span>&#160;            arr_gmin_exp(:) =  arr_gextremes(1:nflds,2)</div><div class="line"><a name="l00598"></a><span class="lineno">  598</span>&#160;</div><div class="line"><a name="l00599"></a><span class="lineno">  599</span>&#160;<span class="comment">! if a field is identically zero, arr_gmin_exp still equals MAXEXPONENT</span></div><div class="line"><a name="l00600"></a><span class="lineno">  600</span>&#160;<span class="comment">!   and arr_gmax_exp still equals MINEXPONENT. In this case, set </span></div><div class="line"><a name="l00601"></a><span class="lineno">  601</span>&#160;<span class="comment">!   arr_gmin_exp = arr_gmax_exp = MINEXPONENT</span></div><div class="line"><a name="l00602"></a><span class="lineno">  602</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00603"></a><span class="lineno">  603</span>&#160;               arr_gmin_exp(ifld) = min(arr_gmax_exp(ifld),arr_gmin_exp(ifld))</div><div class="line"><a name="l00604"></a><span class="lineno">  604</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l00605"></a><span class="lineno">  605</span>&#160;</div><div class="line"><a name="l00606"></a><span class="lineno">  606</span>&#160;<span class="comment">! if requested, return upper bounds on observed maxima </span></div><div class="line"><a name="l00607"></a><span class="lineno">  607</span>&#160;            <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_gbl_max_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00608"></a><span class="lineno">  608</span>&#160;               <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00609"></a><span class="lineno">  609</span>&#160;                  arr_gbl_max_out(ifld) = scale(1.0_r8,arr_gmax_exp(ifld))</div><div class="line"><a name="l00610"></a><span class="lineno">  610</span>&#160;<span class="keywordflow">               enddo</span></div><div class="line"><a name="l00611"></a><span class="lineno">  611</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00612"></a><span class="lineno">  612</span>&#160;</div><div class="line"><a name="l00613"></a><span class="lineno">  613</span>&#160;<span class="comment">! if requested, return max_nsummands before it is redefined</span></div><div class="line"><a name="l00614"></a><span class="lineno">  614</span>&#160;            <span class="keywordflow">if</span> ( <span class="keyword">present</span>( gbl_max_nsummands_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00615"></a><span class="lineno">  615</span>&#160;               gbl_max_nsummands_out = max_nsummands</div><div class="line"><a name="l00616"></a><span class="lineno">  616</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00617"></a><span class="lineno">  617</span>&#160;</div><div class="line"><a name="l00618"></a><span class="lineno">  618</span>&#160;<span class="comment">! determine maximum shift (same as in previous branch, but with calculated</span></div><div class="line"><a name="l00619"></a><span class="lineno">  619</span>&#160;<span class="comment">!  max_nsummands). Shift needs to be small enough that summation does not </span></div><div class="line"><a name="l00620"></a><span class="lineno">  620</span>&#160;<span class="comment">!  exceed maximum number of digits in i8.</span></div><div class="line"><a name="l00621"></a><span class="lineno">  621</span>&#160;</div><div class="line"><a name="l00622"></a><span class="lineno">  622</span>&#160;<span class="comment">! summing within each thread first</span></div><div class="line"><a name="l00623"></a><span class="lineno">  623</span>&#160;            max_nsummands = (max_nsummands/omp_nthreads) + 1</div><div class="line"><a name="l00624"></a><span class="lineno">  624</span>&#160;<span class="comment">! then over threads and tasks</span></div><div class="line"><a name="l00625"></a><span class="lineno">  625</span>&#160;            max_nsummands = max(max_nsummands, tasks*omp_nthreads)</div><div class="line"><a name="l00626"></a><span class="lineno">  626</span>&#160;</div><div class="line"><a name="l00627"></a><span class="lineno">  627</span>&#160;            xmax_nsummands = dble(max_nsummands)</div><div class="line"><a name="l00628"></a><span class="lineno">  628</span>&#160;            arr_max_shift = digits(0_i8) - (exponent(xmax_nsummands) + 1)</div><div class="line"><a name="l00629"></a><span class="lineno">  629</span>&#160;            <span class="keywordflow">if</span> (arr_max_shift &lt; 2) <span class="keywordflow">then</span></div><div class="line"><a name="l00630"></a><span class="lineno">  630</span>&#160;               <span class="keyword">write</span>(<a class="code" href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">nulprt</a>,*) subname,<a class="code" href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">estr</a>,<span class="stringliteral">&#39; number of summands too large for fixed precision algorithm&#39;</span></div><div class="line"><a name="l00631"></a><span class="lineno">  631</span>&#160;               <span class="keyword">call </span><a class="code" href="namespacemod__oasis__sys.html#af1943a0ce0c5765ba8a714379065464f">oasis_abort</a>(file=__file__,line=__line__)</div><div class="line"><a name="l00632"></a><span class="lineno">  632</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00633"></a><span class="lineno">  633</span>&#160;</div><div class="line"><a name="l00634"></a><span class="lineno">  634</span>&#160;<span class="comment">! determine maximum number of levels required for each field</span></div><div class="line"><a name="l00635"></a><span class="lineno">  635</span>&#160;<span class="comment">! ((digits(0_i8) + (arr_gmax_exp(ifld)-arr_gmin_exp(ifld))) / arr_max_shift)</span></div><div class="line"><a name="l00636"></a><span class="lineno">  636</span>&#160;<span class="comment">! + 1 because first truncation probably does not involve a maximal shift</span></div><div class="line"><a name="l00637"></a><span class="lineno">  637</span>&#160;<span class="comment">! + 1 to guarantee that the integer division rounds up (not down)</span></div><div class="line"><a name="l00638"></a><span class="lineno">  638</span>&#160;<span class="comment">! (setting lower bound on max_level*nflds to be 64 to improve OpenMP</span></div><div class="line"><a name="l00639"></a><span class="lineno">  639</span>&#160;<span class="comment">!  performance for loopb in oasis_reprosum_int)</span></div><div class="line"><a name="l00640"></a><span class="lineno">  640</span>&#160;            max_level = (64/nflds) + 1</div><div class="line"><a name="l00641"></a><span class="lineno">  641</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00642"></a><span class="lineno">  642</span>&#160;               max_levels(ifld) = 2 + &amp;</div><div class="line"><a name="l00643"></a><span class="lineno">  643</span>&#160;                ((digits(0_i8) + (arr_gmax_exp(ifld)-arr_gmin_exp(ifld))) &amp;</div><div class="line"><a name="l00644"></a><span class="lineno">  644</span>&#160;                / arr_max_shift)</div><div class="line"><a name="l00645"></a><span class="lineno">  645</span>&#160;               <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_max_levels) .and. (.not. validate) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00646"></a><span class="lineno">  646</span>&#160;<span class="comment">! if validate true, then computation with arr_max_levels failed</span></div><div class="line"><a name="l00647"></a><span class="lineno">  647</span>&#160;<span class="comment">!  previously</span></div><div class="line"><a name="l00648"></a><span class="lineno">  648</span>&#160;                  <span class="keywordflow">if</span> ( arr_max_levels(ifld) &gt; 0 ) <span class="keywordflow">then</span></div><div class="line"><a name="l00649"></a><span class="lineno">  649</span>&#160;                     max_levels(ifld) = &amp;</div><div class="line"><a name="l00650"></a><span class="lineno">  650</span>&#160;                        min(arr_max_levels(ifld),max_levels(ifld))</div><div class="line"><a name="l00651"></a><span class="lineno">  651</span>&#160;<span class="keywordflow">                  endif</span></div><div class="line"><a name="l00652"></a><span class="lineno">  652</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00653"></a><span class="lineno">  653</span>&#160;               <span class="keywordflow">if</span> (max_level &lt; max_levels(ifld)) &amp;</div><div class="line"><a name="l00654"></a><span class="lineno">  654</span>&#160;                  max_level = max_levels(ifld)</div><div class="line"><a name="l00655"></a><span class="lineno">  655</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l00656"></a><span class="lineno">  656</span>&#160;</div><div class="line"><a name="l00657"></a><span class="lineno">  657</span>&#160;<span class="comment">! if requested, return calculated levels</span></div><div class="line"><a name="l00658"></a><span class="lineno">  658</span>&#160;            <span class="keywordflow">if</span> ( <span class="keyword">present</span>(arr_max_levels_out) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00659"></a><span class="lineno">  659</span>&#160;               <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00660"></a><span class="lineno">  660</span>&#160;                  arr_max_levels_out(ifld) = max_levels(ifld)</div><div class="line"><a name="l00661"></a><span class="lineno">  661</span>&#160;<span class="keywordflow">               enddo</span></div><div class="line"><a name="l00662"></a><span class="lineno">  662</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00663"></a><span class="lineno">  663</span>&#160;</div><div class="line"><a name="l00664"></a><span class="lineno">  664</span>&#160;<span class="comment">! calculate sum</span></div><div class="line"><a name="l00665"></a><span class="lineno">  665</span>&#160;            validate = .false.</div><div class="line"><a name="l00666"></a><span class="lineno">  666</span>&#160;            <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a>(arr, arr_gsum, nsummands, dsummands, nflds, &amp;</div><div class="line"><a name="l00667"></a><span class="lineno">  667</span>&#160;                                  arr_max_shift, arr_gmax_exp, max_levels, &amp;</div><div class="line"><a name="l00668"></a><span class="lineno">  668</span>&#160;                                  max_level, validate, recompute, &amp;</div><div class="line"><a name="l00669"></a><span class="lineno">  669</span>&#160;                                  omp_nthreads, mpi_comm)</div><div class="line"><a name="l00670"></a><span class="lineno">  670</span>&#160;</div><div class="line"><a name="l00671"></a><span class="lineno">  671</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l00672"></a><span class="lineno">  672</span>&#160;</div><div class="line"><a name="l00673"></a><span class="lineno">  673</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&#39;oasis_reprosum_int&#39;</span>)</div><div class="line"><a name="l00674"></a><span class="lineno">  674</span>&#160;</div><div class="line"><a name="l00675"></a><span class="lineno">  675</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00676"></a><span class="lineno">  676</span>&#160;</div><div class="line"><a name="l00677"></a><span class="lineno">  677</span>&#160;<span class="comment">! compare fixed and floating point results</span></div><div class="line"><a name="l00678"></a><span class="lineno">  678</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(rel_diff) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00679"></a><span class="lineno">  679</span>&#160;         <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> &gt;= 0.0_r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00680"></a><span class="lineno">  680</span>&#160;</div><div class="line"><a name="l00681"></a><span class="lineno">  681</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keywordflow">then</span></div><div class="line"><a name="l00682"></a><span class="lineno">  682</span>&#160;               <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;oasis_nonreprosum_prebarrier&#39;</span>)</div><div class="line"><a name="l00683"></a><span class="lineno">  683</span>&#160;               <span class="keyword">call </span><a class="code" href="namespacemod__oasis__mpi.html#aaf477847380fbbbc1341520b0b3e66da">oasis_mpi_barrier</a>(mpi_comm,subname)</div><div class="line"><a name="l00684"></a><span class="lineno">  684</span>&#160;               <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a> (<span class="stringliteral">&#39;oasis_nonreprosum_prebarrier&#39;</span>)</div><div class="line"><a name="l00685"></a><span class="lineno">  685</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00686"></a><span class="lineno">  686</span>&#160;</div><div class="line"><a name="l00687"></a><span class="lineno">  687</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;nonrepro_sum&#39;</span>)</div><div class="line"><a name="l00688"></a><span class="lineno">  688</span>&#160;<span class="comment">! record statistic</span></div><div class="line"><a name="l00689"></a><span class="lineno">  689</span>&#160;            nonrepro_sum = 1</div><div class="line"><a name="l00690"></a><span class="lineno">  690</span>&#160;<span class="comment">! compute nonreproducible sum</span></div><div class="line"><a name="l00691"></a><span class="lineno">  691</span>&#160;            arr_lsum(:) = 0._r8</div><div class="line"><a name="l00692"></a><span class="lineno">  692</span>&#160;<span class="comment">!$omp parallel do      &amp;</span></div><div class="line"><a name="l00693"></a><span class="lineno">  693</span>&#160;<span class="comment">!$omp default(shared)  &amp;</span></div><div class="line"><a name="l00694"></a><span class="lineno">  694</span>&#160;<span class="comment">!$omp private(ifld, isum)</span></div><div class="line"><a name="l00695"></a><span class="lineno">  695</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00696"></a><span class="lineno">  696</span>&#160;               <span class="keywordflow">do</span> isum=1,nsummands</div><div class="line"><a name="l00697"></a><span class="lineno">  697</span>&#160;                  arr_lsum(ifld) = arr(isum,ifld) + arr_lsum(ifld)</div><div class="line"><a name="l00698"></a><span class="lineno">  698</span>&#160;<span class="keywordflow">               end do</span></div><div class="line"><a name="l00699"></a><span class="lineno">  699</span>&#160;<span class="keywordflow">            end do</span></div><div class="line"><a name="l00700"></a><span class="lineno">  700</span>&#160;</div><div class="line"><a name="l00701"></a><span class="lineno">  701</span>&#160;            <span class="keyword">call </span>mpi_allreduce (arr_lsum, arr_gsum_fast, nflds, &amp;</div><div class="line"><a name="l00702"></a><span class="lineno">  702</span>&#160;                                mpi_real8, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l00703"></a><span class="lineno">  703</span>&#160;</div><div class="line"><a name="l00704"></a><span class="lineno">  704</span>&#160;            <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&#39;nonrepro_sum&#39;</span>)</div><div class="line"><a name="l00705"></a><span class="lineno">  705</span>&#160;</div><div class="line"><a name="l00706"></a><span class="lineno">  706</span>&#160;<span class="comment">! determine differences</span></div><div class="line"><a name="l00707"></a><span class="lineno">  707</span>&#160;<span class="comment">!$omp parallel do      &amp;</span></div><div class="line"><a name="l00708"></a><span class="lineno">  708</span>&#160;<span class="comment">!$omp default(shared)  &amp;</span></div><div class="line"><a name="l00709"></a><span class="lineno">  709</span>&#160;<span class="comment">!$omp private(ifld, abs_diff)</span></div><div class="line"><a name="l00710"></a><span class="lineno">  710</span>&#160;            <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00711"></a><span class="lineno">  711</span>&#160;               abs_diff = abs(arr_gsum_fast(ifld)-arr_gsum(ifld))</div><div class="line"><a name="l00712"></a><span class="lineno">  712</span>&#160;               <span class="keywordflow">if</span> (abs(arr_gsum(ifld)) &gt; abs_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l00713"></a><span class="lineno">  713</span>&#160;                  rel_diff(1,ifld) = abs_diff/abs(arr_gsum(ifld))</div><div class="line"><a name="l00714"></a><span class="lineno">  714</span>&#160;               <span class="keywordflow">else</span></div><div class="line"><a name="l00715"></a><span class="lineno">  715</span>&#160;                  rel_diff(1,ifld) = abs_diff</div><div class="line"><a name="l00716"></a><span class="lineno">  716</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00717"></a><span class="lineno">  717</span>&#160;               rel_diff(2,ifld) = abs_diff</div><div class="line"><a name="l00718"></a><span class="lineno">  718</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l00719"></a><span class="lineno">  719</span>&#160;         <span class="keywordflow">else</span></div><div class="line"><a name="l00720"></a><span class="lineno">  720</span>&#160;            rel_diff(:,:) = 0.0_r8</div><div class="line"><a name="l00721"></a><span class="lineno">  721</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l00722"></a><span class="lineno">  722</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00723"></a><span class="lineno">  723</span>&#160;</div><div class="line"><a name="l00724"></a><span class="lineno">  724</span>&#160;<span class="comment">! return statistics</span></div><div class="line"><a name="l00725"></a><span class="lineno">  725</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(repro_sum_stats) ) <span class="keywordflow">then</span></div><div class="line"><a name="l00726"></a><span class="lineno">  726</span>&#160;         repro_sum_stats(1) = repro_sum_stats(1) + repro_sum_fast</div><div class="line"><a name="l00727"></a><span class="lineno">  727</span>&#160;         repro_sum_stats(2) = repro_sum_stats(2) + repro_sum_slow</div><div class="line"><a name="l00728"></a><span class="lineno">  728</span>&#160;         repro_sum_stats(3) = repro_sum_stats(3) + repro_sum_both</div><div class="line"><a name="l00729"></a><span class="lineno">  729</span>&#160;         repro_sum_stats(4) = repro_sum_stats(4) + nonrepro_sum</div><div class="line"><a name="l00730"></a><span class="lineno">  730</span>&#160;         repro_sum_stats(5) = repro_sum_stats(5) + gbl_max_red</div><div class="line"><a name="l00731"></a><span class="lineno">  731</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00732"></a><span class="lineno">  732</span>&#160;</div><div class="line"><a name="l00733"></a><span class="lineno">  733</span>&#160;</div><div class="line"><a name="l00734"></a><span class="lineno">  734</span>&#160;<span class="keyword">   end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">oasis_reprosum_calc</a></div><div class="line"><a name="l00735"></a><span class="lineno">  735</span>&#160;</div><div class="line"><a name="l00736"></a><span class="lineno">  736</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l00737"></a><span class="lineno">  737</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l00738"></a><span class="lineno">  738</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00739"></a><span class="lineno">  739</span>&#160;<span class="comment">!&gt; Compute the global sum of each field in &quot;arr&quot; using the indicated </span></div><div class="line"><a name="l00740"></a><span class="lineno">  740</span>&#160;<span class="comment">!&gt; communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l00741"></a><span class="lineno">  741</span>&#160;<span class="comment">!&gt; on a fixed point algorithm. The accuracy of the fixed point algorithm</span></div><div class="line"><a name="l00742"></a><span class="lineno">  742</span>&#160;<span class="comment">!&gt; is controlled by the number of &quot;levels&quot; of integer expansion, the</span></div><div class="line"><a name="l00743"></a><span class="lineno">  743</span>&#160;<span class="comment">!&gt; maximum value of which is specified by max_level. </span></div><div class="line"><a name="l00744"></a><span class="lineno">  744</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l00745"></a><span class="lineno">  745</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00746"></a><span class="lineno">  746</span>&#160;</div><div class="line"><a name="l00747"></a><span class="lineno">  747</span>&#160;<span class="keyword">   subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a> (arr, arr_gsum, nsummands, dsummands, nflds, &amp;</div><div class="line"><a name="l00748"></a><span class="lineno">  748</span>&#160;                                  arr_max_shift, arr_gmax_exp, max_levels,    &amp;</div><div class="line"><a name="l00749"></a><span class="lineno">  749</span>&#160;                                  max_level, validate, recompute,             &amp;</div><div class="line"><a name="l00750"></a><span class="lineno">  750</span>&#160;                                  omp_nthreads, mpi_comm                      )</div><div class="line"><a name="l00751"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">  751</a></span>&#160;</div><div class="line"><a name="l00752"></a><span class="lineno">  752</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00753"></a><span class="lineno">  753</span>&#160;</div><div class="line"><a name="l00754"></a><span class="lineno">  754</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l00755"></a><span class="lineno">  755</span>&#160;</div><div class="line"><a name="l00756"></a><span class="lineno">  756</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nsummands<span class="comment">     !&lt; number of local summands</span></div><div class="line"><a name="l00757"></a><span class="lineno">  757</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: dsummands<span class="comment">     !&lt; declared first dimension</span></div><div class="line"><a name="l00758"></a><span class="lineno">  758</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nflds<span class="comment">         !&lt; number of fields</span></div><div class="line"><a name="l00759"></a><span class="lineno">  759</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: arr_max_shift<span class="comment"> !&lt; maximum safe exponent for </span></div><div class="line"><a name="l00760"></a><span class="lineno">  760</span>&#160;<span class="comment">                                            !&lt;  value &lt; 1 (so that sum </span></div><div class="line"><a name="l00761"></a><span class="lineno">  761</span>&#160;<span class="comment">                                            !&lt;  does not overflow)</span></div><div class="line"><a name="l00762"></a><span class="lineno">  762</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: arr_gmax_exp(nflds)  </div><div class="line"><a name="l00763"></a><span class="lineno">  763</span>&#160;                                            <span class="comment">!&lt; exponents of global maxima</span></div><div class="line"><a name="l00764"></a><span class="lineno">  764</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: max_levels(nflds) </div><div class="line"><a name="l00765"></a><span class="lineno">  765</span>&#160;                                            <span class="comment">!&lt; maximum number of levels </span></div><div class="line"><a name="l00766"></a><span class="lineno">  766</span>&#160;<span class="comment">                                            !&lt;  of integer expansion</span></div><div class="line"><a name="l00767"></a><span class="lineno">  767</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: max_level<span class="comment">     !&lt; maximum value in </span></div><div class="line"><a name="l00768"></a><span class="lineno">  768</span>&#160;<span class="comment">                                            !&lt;  max_levels</span></div><div class="line"><a name="l00769"></a><span class="lineno">  769</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: omp_nthreads<span class="comment">  !&lt; number of OpenMP threads</span></div><div class="line"><a name="l00770"></a><span class="lineno">  770</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: mpi_comm<span class="comment">      !&lt; MPI subcommunicator</span></div><div class="line"><a name="l00771"></a><span class="lineno">  771</span>&#160;</div><div class="line"><a name="l00772"></a><span class="lineno">  772</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span> :: arr(dsummands,nflds) </div><div class="line"><a name="l00773"></a><span class="lineno">  773</span>&#160;                                             <span class="comment">!&lt; input array</span></div><div class="line"><a name="l00774"></a><span class="lineno">  774</span>&#160;</div><div class="line"><a name="l00775"></a><span class="lineno">  775</span>&#160;      <span class="keywordtype">logical</span><span class="keywordtype">intent(in)</span>:: validate</div><div class="line"><a name="l00776"></a><span class="lineno">  776</span>&#160;         <span class="comment">!&lt; flag indicating that accuracy of solution generated from</span></div><div class="line"><a name="l00777"></a><span class="lineno">  777</span>&#160;<span class="comment">         !&lt; arr_gmax_exp and max_levels should be tested</span></div><div class="line"><a name="l00778"></a><span class="lineno">  778</span>&#160;</div><div class="line"><a name="l00779"></a><span class="lineno">  779</span>&#160;      <span class="keywordtype">logical</span><span class="keywordtype">intent(out)</span>:: recompute            </div><div class="line"><a name="l00780"></a><span class="lineno">  780</span>&#160;         <span class="comment">!&lt; flag indicating that either the upper bounds are inaccurate,</span></div><div class="line"><a name="l00781"></a><span class="lineno">  781</span>&#160;<span class="comment">         !&lt;  or max_levels and arr_gmax_exp do not generate accurate </span></div><div class="line"><a name="l00782"></a><span class="lineno">  782</span>&#160;<span class="comment">         !&lt;  enough sums</span></div><div class="line"><a name="l00783"></a><span class="lineno">  783</span>&#160;</div><div class="line"><a name="l00784"></a><span class="lineno">  784</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>:: arr_gsum(nflds)<span class="comment">      !&lt; global means</span></div><div class="line"><a name="l00785"></a><span class="lineno">  785</span>&#160;</div><div class="line"><a name="l00786"></a><span class="lineno">  786</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l00787"></a><span class="lineno">  787</span>&#160;</div><div class="line"><a name="l00788"></a><span class="lineno">  788</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">parameter</span>  :: max_jlevel = &amp;</div><div class="line"><a name="l00789"></a><span class="lineno">  789</span>&#160;                                1 + (digits(0_i8)/digits(0.0_r8))</div><div class="line"><a name="l00790"></a><span class="lineno">  790</span>&#160;</div><div class="line"><a name="l00791"></a><span class="lineno">  791</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_arr_tlsum_level(0:max_level,nflds,omp_nthreads) </div><div class="line"><a name="l00792"></a><span class="lineno">  792</span>&#160;                                   <span class="comment">! integer vector representing local </span></div><div class="line"><a name="l00793"></a><span class="lineno">  793</span>&#160;                                   <span class="comment">!  sum (per thread, per field)</span></div><div class="line"><a name="l00794"></a><span class="lineno">  794</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_arr_lsum_level((max_level+3)*nflds) </div><div class="line"><a name="l00795"></a><span class="lineno">  795</span>&#160;                                   <span class="comment">! integer vector representing local </span></div><div class="line"><a name="l00796"></a><span class="lineno">  796</span>&#160;                                   <span class="comment">!  sum</span></div><div class="line"><a name="l00797"></a><span class="lineno">  797</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_arr_level  <span class="comment">! integer part of summand for current</span></div><div class="line"><a name="l00798"></a><span class="lineno">  798</span>&#160;                                   <span class="comment">!  expansion level</span></div><div class="line"><a name="l00799"></a><span class="lineno">  799</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_arr_gsum_level((max_level+3)*nflds) </div><div class="line"><a name="l00800"></a><span class="lineno">  800</span>&#160;                                   <span class="comment">! integer vector representing global</span></div><div class="line"><a name="l00801"></a><span class="lineno">  801</span>&#160;                                   <span class="comment">!  sum</span></div><div class="line"><a name="l00802"></a><span class="lineno">  802</span>&#160;      <span class="keywordtype">integer(i8)</span> :: IX_8          <span class="comment">! integer representation of current </span></div><div class="line"><a name="l00803"></a><span class="lineno">  803</span>&#160;                                   <span class="comment">!  jlevels of X_8 (&#39;part&#39; of </span></div><div class="line"><a name="l00804"></a><span class="lineno">  804</span>&#160;                                   <span class="comment">!  i8_arr_gsum_level)</span></div><div class="line"><a name="l00805"></a><span class="lineno">  805</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_sign       <span class="comment">! sign global sum</span></div><div class="line"><a name="l00806"></a><span class="lineno">  806</span>&#160;      <span class="keywordtype">integer(i8)</span> :: i8_radix      <span class="comment">! radix for i8 variables</span></div><div class="line"><a name="l00807"></a><span class="lineno">  807</span>&#160;</div><div class="line"><a name="l00808"></a><span class="lineno">  808</span>&#160;      <span class="keywordtype">integer</span> :: max_error(nflds,omp_nthreads)  </div><div class="line"><a name="l00809"></a><span class="lineno">  809</span>&#160;                                   <span class="comment">! accurate upper bound on data?</span></div><div class="line"><a name="l00810"></a><span class="lineno">  810</span>&#160;      <span class="keywordtype">integer</span> :: not_exact(nflds,omp_nthreads)  </div><div class="line"><a name="l00811"></a><span class="lineno">  811</span>&#160;                                   <span class="comment">! max_levels sufficient to </span></div><div class="line"><a name="l00812"></a><span class="lineno">  812</span>&#160;                                   <span class="comment">!  capture all digits?</span></div><div class="line"><a name="l00813"></a><span class="lineno">  813</span>&#160;      <span class="keywordtype">integer</span> :: isum_beg(omp_nthreads), isum_end(omp_nthreads)</div><div class="line"><a name="l00814"></a><span class="lineno">  814</span>&#160;                                   <span class="comment">! range of summand indices for each</span></div><div class="line"><a name="l00815"></a><span class="lineno">  815</span>&#160;                                   <span class="comment">!  OpenMP thread</span></div><div class="line"><a name="l00816"></a><span class="lineno">  816</span>&#160;      <span class="keywordtype">integer</span> :: ifld, isum, ithread</div><div class="line"><a name="l00817"></a><span class="lineno">  817</span>&#160;                                   <span class="comment">! loop variables</span></div><div class="line"><a name="l00818"></a><span class="lineno">  818</span>&#160;      <span class="keywordtype">integer</span> :: arr_exp           <span class="comment">! exponent of summand</span></div><div class="line"><a name="l00819"></a><span class="lineno">  819</span>&#160;      <span class="keywordtype">integer</span> :: arr_shift         <span class="comment">! exponent used to generate integer </span></div><div class="line"><a name="l00820"></a><span class="lineno">  820</span>&#160;                                   <span class="comment">!  for current expansion level</span></div><div class="line"><a name="l00821"></a><span class="lineno">  821</span>&#160;      <span class="keywordtype">integer</span> :: ilevel            <span class="comment">! current integer expansion level</span></div><div class="line"><a name="l00822"></a><span class="lineno">  822</span>&#160;      <span class="keywordtype">integer</span> :: offset(nflds)     <span class="comment">! beginning location in </span></div><div class="line"><a name="l00823"></a><span class="lineno">  823</span>&#160;                                   <span class="comment">!  i8_arr_{g,l}sum_level for integer </span></div><div class="line"><a name="l00824"></a><span class="lineno">  824</span>&#160;                                   <span class="comment">!  expansion of current ifld</span></div><div class="line"><a name="l00825"></a><span class="lineno">  825</span>&#160;      <span class="keywordtype">integer</span> :: voffset           <span class="comment">! modification to offset used to </span></div><div class="line"><a name="l00826"></a><span class="lineno">  826</span>&#160;                                   <span class="comment">!  include validation metrics </span></div><div class="line"><a name="l00827"></a><span class="lineno">  827</span>&#160;      <span class="keywordtype">integer</span> :: ioffset           <span class="comment">! offset(ifld)</span></div><div class="line"><a name="l00828"></a><span class="lineno">  828</span>&#160;      <span class="keywordtype">integer</span> :: jlevel            <span class="comment">! number of floating point &#39;pieces&#39; </span></div><div class="line"><a name="l00829"></a><span class="lineno">  829</span>&#160;                                   <span class="comment">!  extracted from a given i8 integer</span></div><div class="line"><a name="l00830"></a><span class="lineno">  830</span>&#160;      <span class="keywordtype">integer</span> :: ierr              <span class="comment">! MPI error return</span></div><div class="line"><a name="l00831"></a><span class="lineno">  831</span>&#160;      <span class="keywordtype">integer</span> :: LX(max_jlevel)    <span class="comment">! exponent of X_8 (see below)</span></div><div class="line"><a name="l00832"></a><span class="lineno">  832</span>&#160;      <span class="keywordtype">integer</span> :: veclth            <span class="comment">! total length of i8_arr_lsum_level</span></div><div class="line"><a name="l00833"></a><span class="lineno">  833</span>&#160;      <span class="keywordtype">integer</span> :: sum_digits        <span class="comment">! lower bound on number of significant</span></div><div class="line"><a name="l00834"></a><span class="lineno">  834</span>&#160;                                   <span class="comment">!  in integer expansion of sum</span></div><div class="line"><a name="l00835"></a><span class="lineno">  835</span>&#160;      <span class="keywordtype">integer</span> :: curr_exp          <span class="comment">! exponent of partial sum during</span></div><div class="line"><a name="l00836"></a><span class="lineno">  836</span>&#160;                                   <span class="comment">!  reconstruction from integer vector</span></div><div class="line"><a name="l00837"></a><span class="lineno">  837</span>&#160;      <span class="keywordtype">integer</span> :: corr_exp          <span class="comment">! exponent of current summand in</span></div><div class="line"><a name="l00838"></a><span class="lineno">  838</span>&#160;                                   <span class="comment">!  reconstruction from integer vector</span></div><div class="line"><a name="l00839"></a><span class="lineno">  839</span>&#160;</div><div class="line"><a name="l00840"></a><span class="lineno">  840</span>&#160;      <span class="keywordtype">real(r8)</span> :: arr_frac         <span class="comment">! fraction of summand</span></div><div class="line"><a name="l00841"></a><span class="lineno">  841</span>&#160;      <span class="keywordtype">real(r8)</span> :: arr_remainder    <span class="comment">! part of summand remaining after </span></div><div class="line"><a name="l00842"></a><span class="lineno">  842</span>&#160;                                   <span class="comment">!  current level of integer expansion</span></div><div class="line"><a name="l00843"></a><span class="lineno">  843</span>&#160;      <span class="keywordtype">real(r8)</span> :: X_8(max_jlevel)  <span class="comment">! r8 vector representation of current </span></div><div class="line"><a name="l00844"></a><span class="lineno">  844</span>&#160;                                   <span class="comment">!  i8_arr_gsum_level</span></div><div class="line"><a name="l00845"></a><span class="lineno">  845</span>&#160;      <span class="keywordtype">real(r8)</span> :: RX_8             <span class="comment">! r8 representation of difference </span></div><div class="line"><a name="l00846"></a><span class="lineno">  846</span>&#160;                                   <span class="comment">!  between current i8_arr_gsum_level</span></div><div class="line"><a name="l00847"></a><span class="lineno">  847</span>&#160;                                   <span class="comment">!  and current jlevels of X_8 </span></div><div class="line"><a name="l00848"></a><span class="lineno">  848</span>&#160;                                   <span class="comment">!  (== IX_8). Also used in final</span></div><div class="line"><a name="l00849"></a><span class="lineno">  849</span>&#160;                                   <span class="comment">!  scaling step</span></div><div class="line"><a name="l00850"></a><span class="lineno">  850</span>&#160;</div><div class="line"><a name="l00851"></a><span class="lineno">  851</span>&#160;      <span class="keywordtype">logical</span> :: first             <span class="comment">! flag used to indicate that just</span></div><div class="line"><a name="l00852"></a><span class="lineno">  852</span>&#160;                                   <span class="comment">!  beginning reconstruction of sum</span></div><div class="line"><a name="l00853"></a><span class="lineno">  853</span>&#160;                                   <span class="comment">!  from integer vector</span></div><div class="line"><a name="l00854"></a><span class="lineno">  854</span>&#160;</div><div class="line"><a name="l00855"></a><span class="lineno">  855</span>&#160;      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_int)&#39;</span></div><div class="line"><a name="l00856"></a><span class="lineno">  856</span>&#160;</div><div class="line"><a name="l00857"></a><span class="lineno">  857</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00858"></a><span class="lineno">  858</span>&#160;<span class="comment">! Save radix of i8 variables in an i8 variable</span></div><div class="line"><a name="l00859"></a><span class="lineno">  859</span>&#160;      i8_radix = radix(ix_8)</div><div class="line"><a name="l00860"></a><span class="lineno">  860</span>&#160;</div><div class="line"><a name="l00861"></a><span class="lineno">  861</span>&#160;<span class="comment">! If validating upper bounds, reserve space for validation metrics</span></div><div class="line"><a name="l00862"></a><span class="lineno">  862</span>&#160;<span class="comment">! In both cases, reserve an extra level for overflow from the top level</span></div><div class="line"><a name="l00863"></a><span class="lineno">  863</span>&#160;      <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l00864"></a><span class="lineno">  864</span>&#160;        voffset = 3</div><div class="line"><a name="l00865"></a><span class="lineno">  865</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l00866"></a><span class="lineno">  866</span>&#160;        voffset = 1</div><div class="line"><a name="l00867"></a><span class="lineno">  867</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l00868"></a><span class="lineno">  868</span>&#160;</div><div class="line"><a name="l00869"></a><span class="lineno">  869</span>&#160;<span class="comment">! compute offsets for each field</span></div><div class="line"><a name="l00870"></a><span class="lineno">  870</span>&#160;      offset(1) = voffset</div><div class="line"><a name="l00871"></a><span class="lineno">  871</span>&#160;      <span class="keywordflow">do</span> ifld=2,nflds</div><div class="line"><a name="l00872"></a><span class="lineno">  872</span>&#160;         offset(ifld) = offset(ifld-1) &amp;</div><div class="line"><a name="l00873"></a><span class="lineno">  873</span>&#160;                        + (max_levels(ifld-1) + voffset)</div><div class="line"><a name="l00874"></a><span class="lineno">  874</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l00875"></a><span class="lineno">  875</span>&#160;      veclth = offset(nflds) + max_levels(nflds)</div><div class="line"><a name="l00876"></a><span class="lineno">  876</span>&#160;</div><div class="line"><a name="l00877"></a><span class="lineno">  877</span>&#160;<span class="comment">! split summand index range over OpenMP threads</span></div><div class="line"><a name="l00878"></a><span class="lineno">  878</span>&#160;      <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">split_indices</a>(nsummands, omp_nthreads, isum_beg, isum_end)</div><div class="line"><a name="l00879"></a><span class="lineno">  879</span>&#160;</div><div class="line"><a name="l00880"></a><span class="lineno">  880</span>&#160;<span class="comment">! convert local summands to vector of integers and sum</span></div><div class="line"><a name="l00881"></a><span class="lineno">  881</span>&#160;<span class="comment">! (Using scale instead of set_exponent because arr_remainder may not be</span></div><div class="line"><a name="l00882"></a><span class="lineno">  882</span>&#160;<span class="comment">&quot;normal&quot; after level 1 calculation)</span></div><div class="line"><a name="l00883"></a><span class="lineno">  883</span>&#160;      i8_arr_lsum_level(:) = 0_i8</div><div class="line"><a name="l00884"></a><span class="lineno">  884</span>&#160;</div><div class="line"><a name="l00885"></a><span class="lineno">  885</span>&#160;<span class="comment">!$omp parallel do      &amp;</span></div><div class="line"><a name="l00886"></a><span class="lineno">  886</span>&#160;<span class="comment">!$omp default(shared)  &amp;</span></div><div class="line"><a name="l00887"></a><span class="lineno">  887</span>&#160;<span class="comment">!$omp private(ithread, ifld, ioffset, isum, arr_frac, arr_exp, &amp;</span></div><div class="line"><a name="l00888"></a><span class="lineno">  888</span>&#160;<span class="comment">!$omp         arr_shift, ilevel, i8_arr_level, arr_remainder, RX_8, IX_8)</span></div><div class="line"><a name="l00889"></a><span class="lineno">  889</span>&#160;      <span class="keywordflow">do</span> ithread=1,omp_nthreads</div><div class="line"><a name="l00890"></a><span class="lineno">  890</span>&#160;       <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&#39;repro_sum_loopb&#39;</span>)</div><div class="line"><a name="l00891"></a><span class="lineno">  891</span>&#160;       <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00892"></a><span class="lineno">  892</span>&#160;          ioffset = offset(ifld)</div><div class="line"><a name="l00893"></a><span class="lineno">  893</span>&#160;</div><div class="line"><a name="l00894"></a><span class="lineno">  894</span>&#160;          max_error(ifld,ithread) = 0</div><div class="line"><a name="l00895"></a><span class="lineno">  895</span>&#160;          not_exact(ifld,ithread) = 0</div><div class="line"><a name="l00896"></a><span class="lineno">  896</span>&#160;</div><div class="line"><a name="l00897"></a><span class="lineno">  897</span>&#160;          i8_arr_tlsum_level(:,ifld,ithread) = 0_i8</div><div class="line"><a name="l00898"></a><span class="lineno">  898</span>&#160;          <span class="keywordflow">do</span> isum=isum_beg(ithread),isum_end(ithread)</div><div class="line"><a name="l00899"></a><span class="lineno">  899</span>&#160;            arr_remainder = 0.0_r8</div><div class="line"><a name="l00900"></a><span class="lineno">  900</span>&#160;</div><div class="line"><a name="l00901"></a><span class="lineno">  901</span>&#160;            <span class="keywordflow">if</span> (arr(isum,ifld) .ne. 0.0_r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00902"></a><span class="lineno">  902</span>&#160;               arr_exp   = exponent(arr(isum,ifld))</div><div class="line"><a name="l00903"></a><span class="lineno">  903</span>&#160;               arr_frac  = fraction(arr(isum,ifld))</div><div class="line"><a name="l00904"></a><span class="lineno">  904</span>&#160;</div><div class="line"><a name="l00905"></a><span class="lineno">  905</span>&#160;<span class="comment">! test that global maximum upper bound is an upper bound</span></div><div class="line"><a name="l00906"></a><span class="lineno">  906</span>&#160;               <span class="keywordflow">if</span> (arr_exp &gt; arr_gmax_exp(ifld)) <span class="keywordflow">then</span></div><div class="line"><a name="l00907"></a><span class="lineno">  907</span>&#160;                  max_error(ifld,ithread) = 1</div><div class="line"><a name="l00908"></a><span class="lineno">  908</span>&#160;                  <span class="keywordflow">exit</span></div><div class="line"><a name="l00909"></a><span class="lineno">  909</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00910"></a><span class="lineno">  910</span>&#160;</div><div class="line"><a name="l00911"></a><span class="lineno">  911</span>&#160;<span class="comment">! calculate first shift</span></div><div class="line"><a name="l00912"></a><span class="lineno">  912</span>&#160;               arr_shift = arr_max_shift - (arr_gmax_exp(ifld)-arr_exp)</div><div class="line"><a name="l00913"></a><span class="lineno">  913</span>&#160;</div><div class="line"><a name="l00914"></a><span class="lineno">  914</span>&#160;<span class="comment">! determine first (probably) nonzero level (assuming initial fraction is </span></div><div class="line"><a name="l00915"></a><span class="lineno">  915</span>&#160;<span class="comment">&#39;normal&#39; - algorithm still works if this is not true)</span></div><div class="line"><a name="l00916"></a><span class="lineno">  916</span>&#160;<span class="comment">!  NOTE: this is critical; scale will set to zero if min exponent is too small.</span></div><div class="line"><a name="l00917"></a><span class="lineno">  917</span>&#160;               <span class="keywordflow">if</span> (arr_shift &lt; 1) <span class="keywordflow">then</span></div><div class="line"><a name="l00918"></a><span class="lineno">  918</span>&#160;                  ilevel = (1 + (arr_gmax_exp(ifld)-arr_exp))/arr_max_shift</div><div class="line"><a name="l00919"></a><span class="lineno">  919</span>&#160;                  arr_shift = ilevel*arr_max_shift - (arr_gmax_exp(ifld)-arr_exp)</div><div class="line"><a name="l00920"></a><span class="lineno">  920</span>&#160;</div><div class="line"><a name="l00921"></a><span class="lineno">  921</span>&#160;                  <span class="keywordflow">do</span> <span class="keywordflow">while</span> (arr_shift &lt; 1)</div><div class="line"><a name="l00922"></a><span class="lineno">  922</span>&#160;                     arr_shift = arr_shift + arr_max_shift</div><div class="line"><a name="l00923"></a><span class="lineno">  923</span>&#160;                     ilevel = ilevel + 1</div><div class="line"><a name="l00924"></a><span class="lineno">  924</span>&#160;<span class="keywordflow">                  enddo</span></div><div class="line"><a name="l00925"></a><span class="lineno">  925</span>&#160;               <span class="keywordflow">else</span></div><div class="line"><a name="l00926"></a><span class="lineno">  926</span>&#160;                  ilevel = 1</div><div class="line"><a name="l00927"></a><span class="lineno">  927</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00928"></a><span class="lineno">  928</span>&#160;</div><div class="line"><a name="l00929"></a><span class="lineno">  929</span>&#160;               <span class="keywordflow">if</span> (ilevel .le. max_levels(ifld)) <span class="keywordflow">then</span></div><div class="line"><a name="l00930"></a><span class="lineno">  930</span>&#160;<span class="comment">! apply first shift/truncate, add it to the relevant running </span></div><div class="line"><a name="l00931"></a><span class="lineno">  931</span>&#160;<span class="comment">! sum, and calculate the remainder.</span></div><div class="line"><a name="l00932"></a><span class="lineno">  932</span>&#160;                  arr_remainder = scale(arr_frac,arr_shift)</div><div class="line"><a name="l00933"></a><span class="lineno">  933</span>&#160;                  i8_arr_level = int(arr_remainder,i8)</div><div class="line"><a name="l00934"></a><span class="lineno">  934</span>&#160;                  i8_arr_tlsum_level(ilevel,ifld,ithread) = &amp;</div><div class="line"><a name="l00935"></a><span class="lineno">  935</span>&#160;                     i8_arr_tlsum_level(ilevel,ifld,ithread) + i8_arr_level</div><div class="line"><a name="l00936"></a><span class="lineno">  936</span>&#160;                  arr_remainder = arr_remainder - i8_arr_level</div><div class="line"><a name="l00937"></a><span class="lineno">  937</span>&#160;</div><div class="line"><a name="l00938"></a><span class="lineno">  938</span>&#160;<span class="comment">! while the remainder is non-zero, continue to shift, truncate, </span></div><div class="line"><a name="l00939"></a><span class="lineno">  939</span>&#160;<span class="comment">! sum, and calculate new remainder</span></div><div class="line"><a name="l00940"></a><span class="lineno">  940</span>&#160;                  <span class="keywordflow">do</span> <span class="keywordflow">while</span> ((arr_remainder .ne. 0.0_r8) &amp;</div><div class="line"><a name="l00941"></a><span class="lineno">  941</span>&#160;                     .and. (ilevel &lt; max_levels(ifld)))</div><div class="line"><a name="l00942"></a><span class="lineno">  942</span>&#160;                     ilevel = ilevel + 1</div><div class="line"><a name="l00943"></a><span class="lineno">  943</span>&#160;                     arr_remainder = scale(arr_remainder,arr_max_shift)</div><div class="line"><a name="l00944"></a><span class="lineno">  944</span>&#160;                     i8_arr_level = int(arr_remainder,i8)</div><div class="line"><a name="l00945"></a><span class="lineno">  945</span>&#160;                     i8_arr_tlsum_level(ilevel,ifld,ithread) = &amp;</div><div class="line"><a name="l00946"></a><span class="lineno">  946</span>&#160;                        i8_arr_tlsum_level(ilevel,ifld,ithread) + i8_arr_level</div><div class="line"><a name="l00947"></a><span class="lineno">  947</span>&#160;                     arr_remainder = arr_remainder - i8_arr_level</div><div class="line"><a name="l00948"></a><span class="lineno">  948</span>&#160;<span class="keywordflow">                  enddo</span></div><div class="line"><a name="l00949"></a><span class="lineno">  949</span>&#160;</div><div class="line"><a name="l00950"></a><span class="lineno">  950</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l00951"></a><span class="lineno">  951</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00952"></a><span class="lineno">  952</span>&#160;</div><div class="line"><a name="l00953"></a><span class="lineno">  953</span>&#160;            <span class="keywordflow">if</span> (arr_remainder .ne. 0.0_r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00954"></a><span class="lineno">  954</span>&#160;               not_exact(ifld,ithread) = 1</div><div class="line"><a name="l00955"></a><span class="lineno">  955</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l00956"></a><span class="lineno">  956</span>&#160;</div><div class="line"><a name="l00957"></a><span class="lineno">  957</span>&#160;<span class="keywordflow">          enddo</span></div><div class="line"><a name="l00958"></a><span class="lineno">  958</span>&#160;</div><div class="line"><a name="l00959"></a><span class="lineno">  959</span>&#160;<span class="comment">! postprocess integer vector to eliminate potential for overlap in the following</span></div><div class="line"><a name="l00960"></a><span class="lineno">  960</span>&#160;<span class="comment">! sums over threads and processes: if value larger than or equal to </span></div><div class="line"><a name="l00961"></a><span class="lineno">  961</span>&#160;<span class="comment">! (radix(IX_8)**arr_max_shift), add this &#39;overlap&#39; to next larger integer in </span></div><div class="line"><a name="l00962"></a><span class="lineno">  962</span>&#160;<span class="comment">! vector, resulting in nonoverlapping ranges for each component. Note that </span></div><div class="line"><a name="l00963"></a><span class="lineno">  963</span>&#160;<span class="comment">! &quot;ilevel-1==0&quot; corresponds to an extra level used to guarantee that the sums</span></div><div class="line"><a name="l00964"></a><span class="lineno">  964</span>&#160;<span class="comment">! over threads and processes do not overflow for ilevel==1.</span></div><div class="line"><a name="l00965"></a><span class="lineno">  965</span>&#160;          <span class="keywordflow">do</span> ilevel=max_levels(ifld),1,-1</div><div class="line"><a name="l00966"></a><span class="lineno">  966</span>&#160;             rx_8 = i8_arr_tlsum_level(ilevel,ifld,ithread)</div><div class="line"><a name="l00967"></a><span class="lineno">  967</span>&#160;             ix_8 = int(scale(rx_8,-arr_max_shift),i8)</div><div class="line"><a name="l00968"></a><span class="lineno">  968</span>&#160;             <span class="keywordflow">if</span> (ix_8 .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l00969"></a><span class="lineno">  969</span>&#160;                i8_arr_tlsum_level(ilevel-1,ifld,ithread) = &amp;</div><div class="line"><a name="l00970"></a><span class="lineno">  970</span>&#160;                   i8_arr_tlsum_level(ilevel-1,ifld,ithread) + ix_8</div><div class="line"><a name="l00971"></a><span class="lineno">  971</span>&#160;                ix_8 = ix_8*(i8_radix**arr_max_shift)</div><div class="line"><a name="l00972"></a><span class="lineno">  972</span>&#160;                i8_arr_tlsum_level(ilevel,ifld,ithread)   = &amp;</div><div class="line"><a name="l00973"></a><span class="lineno">  973</span>&#160;                   i8_arr_tlsum_level(ilevel,ifld,ithread) - ix_8</div><div class="line"><a name="l00974"></a><span class="lineno">  974</span>&#160;<span class="keywordflow">             endif</span></div><div class="line"><a name="l00975"></a><span class="lineno">  975</span>&#160;<span class="keywordflow">          enddo</span></div><div class="line"><a name="l00976"></a><span class="lineno">  976</span>&#160;<span class="keywordflow">       enddo</span></div><div class="line"><a name="l00977"></a><span class="lineno">  977</span>&#160;       <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&#39;repro_sum_loopb&#39;</span>)</div><div class="line"><a name="l00978"></a><span class="lineno">  978</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l00979"></a><span class="lineno">  979</span>&#160;</div><div class="line"><a name="l00980"></a><span class="lineno">  980</span>&#160;<span class="comment">! sum contributions from different threads</span></div><div class="line"><a name="l00981"></a><span class="lineno">  981</span>&#160;      <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00982"></a><span class="lineno">  982</span>&#160;         ioffset = offset(ifld)</div><div class="line"><a name="l00983"></a><span class="lineno">  983</span>&#160;         <span class="keywordflow">do</span> ithread = 1,omp_nthreads</div><div class="line"><a name="l00984"></a><span class="lineno">  984</span>&#160;            <span class="keywordflow">do</span> ilevel = 0,max_levels(ifld)</div><div class="line"><a name="l00985"></a><span class="lineno">  985</span>&#160;               i8_arr_lsum_level(ioffset+ilevel) = &amp;</div><div class="line"><a name="l00986"></a><span class="lineno">  986</span>&#160;                  i8_arr_lsum_level(ioffset+ilevel) &amp;</div><div class="line"><a name="l00987"></a><span class="lineno">  987</span>&#160;                  + i8_arr_tlsum_level(ilevel,ifld,ithread)</div><div class="line"><a name="l00988"></a><span class="lineno">  988</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l00989"></a><span class="lineno">  989</span>&#160;<span class="keywordflow">         enddo</span></div><div class="line"><a name="l00990"></a><span class="lineno">  990</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l00991"></a><span class="lineno">  991</span>&#160;</div><div class="line"><a name="l00992"></a><span class="lineno">  992</span>&#160;<span class="comment">! record if upper bound was inaccurate or if level expansion stopped</span></div><div class="line"><a name="l00993"></a><span class="lineno">  993</span>&#160;<span class="comment">! before full accuracy was achieved</span></div><div class="line"><a name="l00994"></a><span class="lineno">  994</span>&#160;      <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l00995"></a><span class="lineno">  995</span>&#160;         <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00996"></a><span class="lineno">  996</span>&#160;            ioffset = offset(ifld)</div><div class="line"><a name="l00997"></a><span class="lineno">  997</span>&#160;            i8_arr_lsum_level(ioffset-voffset+1) = maxval(max_error(ifld,:))</div><div class="line"><a name="l00998"></a><span class="lineno">  998</span>&#160;            i8_arr_lsum_level(ioffset-voffset+2) = maxval(not_exact(ifld,:))</div><div class="line"><a name="l00999"></a><span class="lineno">  999</span>&#160;<span class="keywordflow">         enddo</span></div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span>&#160;</div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span>&#160;<span class="comment">! sum integer vector element-wise</span></div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span>&#160;<span class="preprocessor">#if ( defined noI8 )</span></div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span>&#160;<span class="preprocessor"></span>     <span class="comment">! Workaround for when i8 is not supported.</span></div><div class="line"><a name="l01005"></a><span class="lineno"> 1005</span>&#160;      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&quot;repro_sum_allr_i4&quot;</span>)</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>&#160;      <span class="keyword">call </span>mpi_allreduce (i8_arr_lsum_level, i8_arr_gsum_level, &amp; </div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>&#160;                          veclth, mpi_integer, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</span>&#160;      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&quot;repro_sum_allr_i4&quot;</span>)</div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span>&#160;<span class="preprocessor">#else</span></div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span>&#160;<span class="preprocessor"></span>      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">&quot;repro_sum_allr_i8&quot;</span>)</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>&#160;      <span class="keyword">call </span>mpi_allreduce (i8_arr_lsum_level, i8_arr_gsum_level, &amp; </div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>&#160;                          veclth, mpi_integer8, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</span>&#160;      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">detailed_timing</a>) <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">&quot;repro_sum_allr_i8&quot;</span>)</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span>&#160;<span class="preprocessor">#endif</span></div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span>&#160;<span class="preprocessor"></span></div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span>&#160;<span class="comment">! Construct global sum from integer vector representation:</span></div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span>&#160;<span class="comment">!  1) arr_max_shift is the shift applied to fraction(arr_gmax) .</span></div><div class="line"><a name="l01018"></a><span class="lineno"> 1018</span>&#160;<span class="comment">!   When shifting back, need to &quot;add back in&quot; true arr_gmax exponent. This was</span></div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span>&#160;<span class="comment">!   removed implicitly by working only with the fraction .</span></div><div class="line"><a name="l01020"></a><span class="lineno"> 1020</span>&#160;<span class="comment">!  2) want to add levels into sum in reverse order (smallest to largest). However,</span></div><div class="line"><a name="l01021"></a><span class="lineno"> 1021</span>&#160;<span class="comment">!   even this can generate floating point rounding errors if signs of integers</span></div><div class="line"><a name="l01022"></a><span class="lineno"> 1022</span>&#160;<span class="comment">!   alternate. To avoid this, do some arithmetic with integer vectors so that all</span></div><div class="line"><a name="l01023"></a><span class="lineno"> 1023</span>&#160;<span class="comment">!   components have the same sign. This should keep relative difference between</span></div><div class="line"><a name="l01024"></a><span class="lineno"> 1024</span>&#160;<span class="comment">!   using different integer sizes (e.g. i8 and i4) to machine epsilon</span></div><div class="line"><a name="l01025"></a><span class="lineno"> 1025</span>&#160;<span class="comment">!  3) assignment to X_8 will usually lose accuracy since maximum integer </span></div><div class="line"><a name="l01026"></a><span class="lineno"> 1026</span>&#160;<span class="comment">!   size is greater than the max number of &#39;digits&#39; in r8 value (if xmax_nsummands </span></div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span>&#160;<span class="comment">!   correction not very large). Calculate remainder and add in first (since </span></div><div class="line"><a name="l01028"></a><span class="lineno"> 1028</span>&#160;<span class="comment">!   smaller). One correction is sufficient for r8 (53 digits) and i8 (63 digits). </span></div><div class="line"><a name="l01029"></a><span class="lineno"> 1029</span>&#160;<span class="comment">!   For r4 (24 digits) may need to correct twice. Code is written in a general</span></div><div class="line"><a name="l01030"></a><span class="lineno"> 1030</span>&#160;<span class="comment">!   fashion, to work no matter how many corrections are necessary (assuming</span></div><div class="line"><a name="l01031"></a><span class="lineno"> 1031</span>&#160;<span class="comment">!   max_jlevel parameter calculation is correct).</span></div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span>&#160;</div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span>&#160;      recompute = .false.</div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span>&#160;      <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>&#160;         arr_gsum(ifld) = 0.0_r8</div><div class="line"><a name="l01036"></a><span class="lineno"> 1036</span>&#160;         ioffset = offset(ifld)</div><div class="line"><a name="l01037"></a><span class="lineno"> 1037</span>&#160;</div><div class="line"><a name="l01038"></a><span class="lineno"> 1038</span>&#160;<span class="comment">! if validate is .true., test whether the summand upper bound</span></div><div class="line"><a name="l01039"></a><span class="lineno"> 1039</span>&#160;<span class="comment">!  was exceeded on any of the processes</span></div><div class="line"><a name="l01040"></a><span class="lineno"> 1040</span>&#160;         <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l01041"></a><span class="lineno"> 1041</span>&#160;            <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset-voffset+1) .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01042"></a><span class="lineno"> 1042</span>&#160;               recompute = .true.</div><div class="line"><a name="l01043"></a><span class="lineno"> 1043</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01044"></a><span class="lineno"> 1044</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l01045"></a><span class="lineno"> 1045</span>&#160;</div><div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>&#160;         <span class="keywordflow">if</span> (.not. recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l01047"></a><span class="lineno"> 1047</span>&#160;</div><div class="line"><a name="l01048"></a><span class="lineno"> 1048</span>&#160;<span class="comment">! preprocess integer vector:</span></div><div class="line"><a name="l01049"></a><span class="lineno"> 1049</span>&#160;<span class="comment">!  a) if value larger than or equal to (radix(IX_8)**arr_max_shift), add this &#39;overlap&#39; </span></div><div class="line"><a name="l01050"></a><span class="lineno"> 1050</span>&#160;<span class="comment">!     to next larger integer in vector, resulting in nonoverlapping ranges for each</span></div><div class="line"><a name="l01051"></a><span class="lineno"> 1051</span>&#160;<span class="comment">!     component. Note that have &quot;ilevel-1=0&quot; level here as described above.</span></div><div class="line"><a name="l01052"></a><span class="lineno"> 1052</span>&#160;           <span class="keywordflow">do</span> ilevel=max_levels(ifld),1,-1</div><div class="line"><a name="l01053"></a><span class="lineno"> 1053</span>&#160;             rx_8 = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>&#160;             ix_8 = int(scale(rx_8,-arr_max_shift),i8)</div><div class="line"><a name="l01055"></a><span class="lineno"> 1055</span>&#160;             <span class="keywordflow">if</span> (ix_8 .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01056"></a><span class="lineno"> 1056</span>&#160;               i8_arr_gsum_level(ioffset+ilevel-1) = i8_arr_gsum_level(ioffset+ilevel-1) &amp;</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>&#160;                                                   + ix_8</div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>&#160;               ix_8 = ix_8*(i8_radix**arr_max_shift)</div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span>&#160;               i8_arr_gsum_level(ioffset+ilevel)   = i8_arr_gsum_level(ioffset+ilevel)   &amp;</div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span>&#160;                                                   - ix_8</div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span>&#160;<span class="keywordflow">             endif</span></div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span>&#160;<span class="keywordflow">           enddo</span></div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span>&#160;<span class="comment">!  b) subtract +/- 1 from larger and add +/- 1 to smaller when necessary </span></div><div class="line"><a name="l01064"></a><span class="lineno"> 1064</span>&#160;<span class="comment">!     so that all vector components have the same sign (eliminating loss </span></div><div class="line"><a name="l01065"></a><span class="lineno"> 1065</span>&#160;<span class="comment">!     of accuracy arising from difference of large values when </span></div><div class="line"><a name="l01066"></a><span class="lineno"> 1066</span>&#160;<span class="comment">!     reconstructing r8 sum from integer vector)</span></div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span>&#160;           ilevel = 0</div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span>&#160;           <span class="keywordflow">do</span> <span class="keywordflow">while</span> ((i8_arr_gsum_level(ioffset+ilevel) .eq. 0_i8) &amp;</div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span>&#160;                     .and. (ilevel &lt; max_levels(ifld)))</div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span>&#160;             ilevel = ilevel + 1</div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span>&#160;<span class="keywordflow">           enddo</span></div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span>&#160;<span class="comment">!</span></div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>&#160;           <span class="keywordflow">if</span> (ilevel &lt; max_levels(ifld)) <span class="keywordflow">then</span></div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span>&#160;              <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset+ilevel) &gt; 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span>&#160;                 i8_sign = 1_i8</div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span>&#160;              <span class="keywordflow">else</span></div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span>&#160;                 i8_sign = -1_i8</div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span>&#160;<span class="keywordflow">              endif</span></div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span>&#160;              <span class="keywordflow">do</span> jlevel=ilevel,max_levels(ifld)-1</div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span>&#160;                 <span class="keywordflow">if</span> (sign(1_i8,i8_arr_gsum_level(ioffset+jlevel)) &amp;</div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>&#160;                     .ne. sign(1_i8,i8_arr_gsum_level(ioffset+jlevel+1))) <span class="keywordflow">then</span></div><div class="line"><a name="l01082"></a><span class="lineno"> 1082</span>&#160;                    i8_arr_gsum_level(ioffset+jlevel)   = i8_arr_gsum_level(ioffset+jlevel) &amp;</div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span>&#160;                                                        - i8_sign</div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>&#160;                    i8_arr_gsum_level(ioffset+jlevel+1) = i8_arr_gsum_level(ioffset+jlevel+1) &amp;</div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span>&#160;                                                        + i8_sign*(i8_radix**arr_max_shift)</div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span>&#160;<span class="keywordflow">                 endif</span></div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</span>&#160;<span class="keywordflow">              enddo</span>                </div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span>&#160;</div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span>&#160;<span class="comment">! start with maximum shift, and work up to larger values</span></div><div class="line"><a name="l01091"></a><span class="lineno"> 1091</span>&#160;            arr_shift = arr_gmax_exp(ifld) &amp;</div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span>&#160;                        - max_levels(ifld)*arr_max_shift</div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span>&#160;            curr_exp = 0</div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span>&#160;            first = .true.</div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>&#160;            <span class="keywordflow">do</span> ilevel=max_levels(ifld),0,-1</div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span>&#160;</div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>&#160;               <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset+ilevel) .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01098"></a><span class="lineno"> 1098</span>&#160;                  jlevel = 1</div><div class="line"><a name="l01099"></a><span class="lineno"> 1099</span>&#160;</div><div class="line"><a name="l01100"></a><span class="lineno"> 1100</span>&#160;<span class="comment">! r8 representation of higher order bits in integer </span></div><div class="line"><a name="l01101"></a><span class="lineno"> 1101</span>&#160;                  x_8(jlevel) = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01102"></a><span class="lineno"> 1102</span>&#160;                  lx(jlevel)  = exponent(x_8(jlevel))</div><div class="line"><a name="l01103"></a><span class="lineno"> 1103</span>&#160;</div><div class="line"><a name="l01104"></a><span class="lineno"> 1104</span>&#160;<span class="comment">! calculate remainder</span></div><div class="line"><a name="l01105"></a><span class="lineno"> 1105</span>&#160;                  ix_8 = int(x_8(jlevel),i8)</div><div class="line"><a name="l01106"></a><span class="lineno"> 1106</span>&#160;                  rx_8 = (i8_arr_gsum_level(ioffset+ilevel) - ix_8)</div><div class="line"><a name="l01107"></a><span class="lineno"> 1107</span>&#160;</div><div class="line"><a name="l01108"></a><span class="lineno"> 1108</span>&#160;<span class="comment">! repeat using remainder</span></div><div class="line"><a name="l01109"></a><span class="lineno"> 1109</span>&#160;                  <span class="keywordflow">do</span> <span class="keywordflow">while</span> (rx_8 .ne. 0.0_r8)</div><div class="line"><a name="l01110"></a><span class="lineno"> 1110</span>&#160;                     jlevel = jlevel + 1</div><div class="line"><a name="l01111"></a><span class="lineno"> 1111</span>&#160;                     x_8(jlevel) = rx_8</div><div class="line"><a name="l01112"></a><span class="lineno"> 1112</span>&#160;                     lx(jlevel) = exponent(rx_8)</div><div class="line"><a name="l01113"></a><span class="lineno"> 1113</span>&#160;                     ix_8 = ix_8 + int(rx_8,i8)</div><div class="line"><a name="l01114"></a><span class="lineno"> 1114</span>&#160;                     rx_8 = (i8_arr_gsum_level(ioffset+ilevel) - ix_8)</div><div class="line"><a name="l01115"></a><span class="lineno"> 1115</span>&#160;<span class="keywordflow">                  enddo</span></div><div class="line"><a name="l01116"></a><span class="lineno"> 1116</span>&#160;</div><div class="line"><a name="l01117"></a><span class="lineno"> 1117</span>&#160;<span class="comment">! add in contributions, smaller to larger, rescaling for each</span></div><div class="line"><a name="l01118"></a><span class="lineno"> 1118</span>&#160;<span class="comment">! addition to guarantee that exponent of working summand is always </span></div><div class="line"><a name="l01119"></a><span class="lineno"> 1119</span>&#160;<span class="comment">! larger than minexponent</span></div><div class="line"><a name="l01120"></a><span class="lineno"> 1120</span>&#160;                  <span class="keywordflow">do</span> <span class="keywordflow">while</span> (jlevel &gt; 0)</div><div class="line"><a name="l01121"></a><span class="lineno"> 1121</span>&#160;                     <span class="keywordflow">if</span> (first) <span class="keywordflow">then</span></div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span>&#160;                        curr_exp = lx(jlevel) + arr_shift</div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>&#160;                        arr_gsum(ifld) = fraction(x_8(jlevel))</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>&#160;                        first = .false.</div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>&#160;                     <span class="keywordflow">else</span></div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span>&#160;                        corr_exp = curr_exp - (lx(jlevel) + arr_shift)</div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>&#160;                        arr_gsum(ifld) = fraction(x_8(jlevel)) &amp;</div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>&#160;                                       + scale(arr_gsum(ifld),corr_exp)</div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>&#160;                        curr_exp = lx(jlevel) + arr_shift</div><div class="line"><a name="l01130"></a><span class="lineno"> 1130</span>&#160;<span class="keywordflow">                     endif</span></div><div class="line"><a name="l01131"></a><span class="lineno"> 1131</span>&#160;                     jlevel = jlevel - 1</div><div class="line"><a name="l01132"></a><span class="lineno"> 1132</span>&#160;<span class="keywordflow">                  enddo</span></div><div class="line"><a name="l01133"></a><span class="lineno"> 1133</span>&#160;</div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span>&#160;</div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>&#160;               arr_shift = arr_shift + arr_max_shift</div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span>&#160;<span class="keywordflow">            enddo</span></div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span>&#160;</div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span>&#160;<span class="comment">! apply final exponent correction, scaling first if exponent is too small</span></div><div class="line"><a name="l01140"></a><span class="lineno"> 1140</span>&#160;<span class="comment">! to apply directly</span></div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>&#160;            corr_exp = curr_exp + exponent(arr_gsum(ifld))</div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span>&#160;            <span class="keywordflow">if</span> (corr_exp .ge. minexponent(1._r8)) <span class="keywordflow">then</span> </div><div class="line"><a name="l01143"></a><span class="lineno"> 1143</span>&#160;               arr_gsum(ifld) = set_exponent(arr_gsum(ifld),corr_exp)</div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>&#160;            <span class="keywordflow">else</span></div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>&#160;               rx_8 = set_exponent(arr_gsum(ifld), &amp;</div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span>&#160;                                   corr_exp-minexponent(1._r8))</div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span>&#160;               arr_gsum(ifld) = scale(rx_8,minexponent(1._r8))</div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span>&#160;</div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span>&#160;<span class="comment">! if validate is .true. and some precision lost, test whether &#39;too much&#39; </span></div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span>&#160;<span class="comment">!  was lost, due to too loose an upper bound, too stringent a limit on number</span></div><div class="line"><a name="l01152"></a><span class="lineno"> 1152</span>&#160;<span class="comment">!  of levels of expansion, cancellation, .... Calculated by comparing lower </span></div><div class="line"><a name="l01153"></a><span class="lineno"> 1153</span>&#160;<span class="comment">!  bound on number of sigificant digits with number of digits in 1.0_r8 .</span></div><div class="line"><a name="l01154"></a><span class="lineno"> 1154</span>&#160;            <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>&#160;               <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset-voffset+2) .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01156"></a><span class="lineno"> 1156</span>&#160;</div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span>&#160;<span class="comment">! find first nonzero level and use exponent for this level, then assume all</span></div><div class="line"><a name="l01158"></a><span class="lineno"> 1158</span>&#160;<span class="comment">! subsequent levels contribute arr_max_shift digits.</span></div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span>&#160;                  sum_digits = 0</div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span>&#160;                  <span class="keywordflow">do</span> ilevel=0,max_levels(ifld)</div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>&#160;                     <span class="keywordflow">if</span> (sum_digits .eq. 0) <span class="keywordflow">then</span></div><div class="line"><a name="l01162"></a><span class="lineno"> 1162</span>&#160;                        <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset+ilevel) .ne. 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01163"></a><span class="lineno"> 1163</span>&#160;                           x_8(1) = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>&#160;                           lx(1)  = exponent(x_8(1))</div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>&#160;                           sum_digits = lx(1)</div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span>&#160;<span class="keywordflow">                        endif</span></div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>&#160;                     <span class="keywordflow">else</span></div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>&#160;                        sum_digits = sum_digits + arr_max_shift</div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span>&#160;<span class="keywordflow">                     endif</span></div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span>&#160;<span class="keywordflow">                  enddo</span></div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span>&#160;</div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>&#160;                  <span class="keywordflow">if</span> (sum_digits &lt; digits(1.0_r8)) <span class="keywordflow">then</span></div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>&#160;                     recompute = .true.</div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span>&#160;<span class="keywordflow">                  endif</span></div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span>&#160;<span class="keywordflow">               endif</span></div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span>&#160;</div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span>&#160;</div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span>&#160;</div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span>&#160;</div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span>&#160;<span class="keyword">   end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a></div><div class="line"><a name="l01184"></a><span class="lineno"> 1184</span>&#160;</div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span>&#160;<span class="comment">!&gt; Test whether distributed sum exceeds tolerance and print out a</span></div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span>&#160;<span class="comment">!&gt; warning message.</span></div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span>&#160;</div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span>&#160;<span class="keyword">   logical </span><span class="keyword">function </span><a class="code" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">oasis_reprosum_tolexceeded</a> (name, nflds, master, &amp;</div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>&#160;                                                logunit, rel_diff    )</div><div class="line"><a name="l01195"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08"> 1195</a></span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span>&#160;</div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span>&#160;</div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>&#160;      <span class="keywordtype">character(len=*)</span>, <span class="keywordtype">intent(in)</span> :: name<span class="comment">    !&lt; distributed sum identifier</span></div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nflds<span class="comment">           !&lt; number of fields</span></div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>&#160;      <span class="keywordtype">logical</span><span class="keywordtype">intent(in)</span> :: master<span class="comment">          !&lt; process that will write</span></div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span>&#160;<span class="comment">                                              !&lt;  warning messages?</span></div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">optional</span>, <span class="keywordtype">intent(in)</span> :: logunit<span class="comment">!&lt; unit warning messages </span></div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span>&#160;<span class="comment">                                              !&lt;  written to</span></div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span> :: rel_diff(2,nflds)</div><div class="line"><a name="l01206"></a><span class="lineno"> 1206</span>&#160;                                              <span class="comment">!&lt; relative and absolute</span></div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span>&#160;<span class="comment">                                              !&lt;  differences between fixed </span></div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span>&#160;<span class="comment">                                              !&lt;  and floating point sums</span></div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span>&#160;</div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span>&#160;</div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>&#160;      <span class="keywordtype">integer</span>  :: llogunit                    <span class="comment">! local log unit</span></div><div class="line"><a name="l01213"></a><span class="lineno"> 1213</span>&#160;      <span class="keywordtype">integer</span>  :: ifld                        <span class="comment">! field index</span></div><div class="line"><a name="l01214"></a><span class="lineno"> 1214</span>&#160;      <span class="keywordtype">integer</span>  :: exceeds_limit               <span class="comment">! number of fields whose</span></div><div class="line"><a name="l01215"></a><span class="lineno"> 1215</span>&#160;                                              <span class="comment">!  sum exceeds tolerance</span></div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span>&#160;      <span class="keywordtype">real(r8)</span> :: max_rel_diff                <span class="comment">! maximum relative difference</span></div><div class="line"><a name="l01217"></a><span class="lineno"> 1217</span>&#160;      <span class="keywordtype">integer</span>  :: max_rel_diff_idx            <span class="comment">! field index for max. rel. diff.</span></div><div class="line"><a name="l01218"></a><span class="lineno"> 1218</span>&#160;      <span class="keywordtype">real(r8)</span> :: max_abs_diff                <span class="comment">! maximum absolute difference</span></div><div class="line"><a name="l01219"></a><span class="lineno"> 1219</span>&#160;      <span class="keywordtype">integer</span>  :: max_abs_diff_idx            <span class="comment">! field index for max. abs. diff.</span></div><div class="line"><a name="l01220"></a><span class="lineno"> 1220</span>&#160;      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_tolExceeded)&#39;</span></div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span>&#160;</div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span>&#160;</div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>&#160;      <a class="code" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">oasis_reprosum_tolexceeded</a> = .false.</div><div class="line"><a name="l01225"></a><span class="lineno"> 1225</span>&#160;      <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> &lt; 0.0_r8) <span class="keywordflow">return</span></div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span>&#160;</div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>&#160;      <span class="keywordflow">if</span> ( <span class="keyword">present</span>(logunit) ) <span class="keywordflow">then</span></div><div class="line"><a name="l01228"></a><span class="lineno"> 1228</span>&#160;         llogunit = logunit</div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>&#160;      <span class="keywordflow">else</span></div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>&#160;         llogunit = <a class="code" href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">nulprt</a></div><div class="line"><a name="l01231"></a><span class="lineno"> 1231</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span>&#160;</div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span>&#160;      <span class="comment">! check that &quot;fast&quot; reproducible sum is accurate enough.</span></div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>&#160;      exceeds_limit = 0</div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>&#160;      max_rel_diff = 0.0_r8</div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>&#160;      max_abs_diff = 0.0_r8</div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>&#160;      <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span>&#160;         <span class="keywordflow">if</span> (rel_diff(1,ifld) &gt; <a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a>) <span class="keywordflow">then</span></div><div class="line"><a name="l01239"></a><span class="lineno"> 1239</span>&#160;            exceeds_limit = exceeds_limit + 1</div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span>&#160;            <span class="keywordflow">if</span> (rel_diff(1,ifld) &gt; max_rel_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span>&#160;               max_rel_diff = rel_diff(1,ifld)</div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>&#160;               max_rel_diff_idx = ifld</div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span>&#160;            <span class="keywordflow">if</span> (rel_diff(2,ifld) &gt; max_abs_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>&#160;               max_abs_diff = rel_diff(2,ifld)</div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span>&#160;               max_abs_diff_idx = ifld</div><div class="line"><a name="l01247"></a><span class="lineno"> 1247</span>&#160;<span class="keywordflow">            endif</span></div><div class="line"><a name="l01248"></a><span class="lineno"> 1248</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l01249"></a><span class="lineno"> 1249</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l01250"></a><span class="lineno"> 1250</span>&#160;</div><div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>&#160;      <span class="keywordflow">if</span> (exceeds_limit &gt; 0) <span class="keywordflow">then</span></div><div class="line"><a name="l01252"></a><span class="lineno"> 1252</span>&#160;         <span class="keywordflow">if</span> (master) <span class="keywordflow">then</span></div><div class="line"><a name="l01253"></a><span class="lineno"> 1253</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname,trim(name), &amp;</div><div class="line"><a name="l01254"></a><span class="lineno"> 1254</span>&#160;                            <span class="stringliteral">&#39;: difference in fixed and floating point sums &#39;</span>, &amp;</div><div class="line"><a name="l01255"></a><span class="lineno"> 1255</span>&#160;                            <span class="stringliteral">&#39; exceeds tolerance in &#39;</span>, exceeds_limit, &amp;</div><div class="line"><a name="l01256"></a><span class="lineno"> 1256</span>&#160;                            <span class="stringliteral">&#39; fields.&#39;</span></div><div class="line"><a name="l01257"></a><span class="lineno"> 1257</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname,<span class="stringliteral">&#39;  Maximum relative diff: (rel)&#39;</span>, &amp;</div><div class="line"><a name="l01258"></a><span class="lineno"> 1258</span>&#160;                            rel_diff(1,max_rel_diff_idx), <span class="stringliteral">&#39; (abs) &#39;</span>, &amp;</div><div class="line"><a name="l01259"></a><span class="lineno"> 1259</span>&#160;                            rel_diff(2,max_rel_diff_idx)</div><div class="line"><a name="l01260"></a><span class="lineno"> 1260</span>&#160;            <span class="keyword">write</span>(llogunit,*) subname,<span class="stringliteral">&#39;  Maximum absolute diff: (rel)&#39;</span>, &amp;</div><div class="line"><a name="l01261"></a><span class="lineno"> 1261</span>&#160;                            rel_diff(1,max_abs_diff_idx), <span class="stringliteral">&#39; (abs) &#39;</span>, &amp;</div><div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>&#160;                            rel_diff(2,max_abs_diff_idx)</div><div class="line"><a name="l01263"></a><span class="lineno"> 1263</span>&#160;<span class="keywordflow">         endif</span></div><div class="line"><a name="l01264"></a><span class="lineno"> 1264</span>&#160;         <a class="code" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">oasis_reprosum_tolexceeded</a> = .true.</div><div class="line"><a name="l01265"></a><span class="lineno"> 1265</span>&#160;<span class="keywordflow">       endif</span></div><div class="line"><a name="l01266"></a><span class="lineno"> 1266</span>&#160;</div><div class="line"><a name="l01267"></a><span class="lineno"> 1267</span>&#160;</div><div class="line"><a name="l01268"></a><span class="lineno"> 1268</span>&#160;<span class="keyword">   end function </span><a class="code" href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">oasis_reprosum_tolexceeded</a></div><div class="line"><a name="l01269"></a><span class="lineno"> 1269</span>&#160;</div><div class="line"><a name="l01270"></a><span class="lineno"> 1270</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l01271"></a><span class="lineno"> 1271</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01272"></a><span class="lineno"> 1272</span>&#160;<span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l01273"></a><span class="lineno"> 1273</span>&#160;<span class="comment">!&gt; Compute the global sum of each field in &quot;arr&quot; using the indicated </span></div><div class="line"><a name="l01274"></a><span class="lineno"> 1274</span>&#160;<span class="comment">!&gt; communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l01275"></a><span class="lineno"> 1275</span>&#160;<span class="comment">!&gt; on He and Ding&#39;s implementation of the double-double algorithm.</span></div><div class="line"><a name="l01276"></a><span class="lineno"> 1276</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01277"></a><span class="lineno"> 1277</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01278"></a><span class="lineno"> 1278</span>&#160;</div><div class="line"><a name="l01279"></a><span class="lineno"> 1279</span>&#160;<span class="keyword">   subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">oasis_reprosum_ddpdd</a> (arr, arr_gsum, nsummands, dsummands,  &amp;</div><div class="line"><a name="l01280"></a><span class="lineno"> 1280</span>&#160;                                    nflds, mpi_comm                       )</div><div class="line"><a name="l01281"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102"> 1281</a></span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01282"></a><span class="lineno"> 1282</span>&#160;</div><div class="line"><a name="l01283"></a><span class="lineno"> 1283</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l01284"></a><span class="lineno"> 1284</span>&#160;</div><div class="line"><a name="l01285"></a><span class="lineno"> 1285</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nsummands<span class="comment">  !&lt; number of local summands</span></div><div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: dsummands<span class="comment">  !&lt; declared first dimension</span></div><div class="line"><a name="l01287"></a><span class="lineno"> 1287</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: nflds<span class="comment">      !&lt; number of fields</span></div><div class="line"><a name="l01288"></a><span class="lineno"> 1288</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(in)</span> :: arr(dsummands,nflds) </div><div class="line"><a name="l01289"></a><span class="lineno"> 1289</span>&#160;                                         <span class="comment">!&lt; input array</span></div><div class="line"><a name="l01290"></a><span class="lineno"> 1290</span>&#160;      <span class="keywordtype">integer</span><span class="keywordtype">intent(in)</span> :: mpi_comm<span class="comment">   !&lt; MPI subcommunicator</span></div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span>&#160;</div><div class="line"><a name="l01292"></a><span class="lineno"> 1292</span>&#160;      <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>:: arr_gsum(nflds)      </div><div class="line"><a name="l01293"></a><span class="lineno"> 1293</span>&#160;                                         <span class="comment">!&lt; global sums</span></div><div class="line"><a name="l01294"></a><span class="lineno"> 1294</span>&#160;</div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span>&#160;</div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span>&#160;</div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>&#160;      <span class="keywordtype">integer</span> :: old_cw                  <span class="comment">! for x86 processors, save</span></div><div class="line"><a name="l01299"></a><span class="lineno"> 1299</span>&#160;                                         <span class="comment">!  current arithmetic mode</span></div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span>&#160;      <span class="keywordtype">integer</span> :: ifld, isum              <span class="comment">! loop variables</span></div><div class="line"><a name="l01301"></a><span class="lineno"> 1301</span>&#160;      <span class="keywordtype">integer</span> :: ierr                    <span class="comment">! MPI error return</span></div><div class="line"><a name="l01302"></a><span class="lineno"> 1302</span>&#160;</div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>&#160;      <span class="keywordtype">real(r8)</span>    :: e, t1, t2           <span class="comment">! temporaries</span></div><div class="line"><a name="l01304"></a><span class="lineno"> 1304</span>&#160;      <span class="keywordtype">complex(r8)</span> :: arr_lsum_dd(nflds)  <span class="comment">! local sums (in double-double</span></div><div class="line"><a name="l01305"></a><span class="lineno"> 1305</span>&#160;                                         <span class="comment">!  format)</span></div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>&#160;      <span class="keywordtype">complex(r8)</span> :: arr_gsum_dd(nflds)  <span class="comment">! global sums (in double-double</span></div><div class="line"><a name="l01307"></a><span class="lineno"> 1307</span>&#160;                                         <span class="comment">!  format)</span></div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span>&#160;</div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">save</span> :: mpi_sumdd</div><div class="line"><a name="l01310"></a><span class="lineno"> 1310</span>&#160;      <span class="keywordtype">logical</span>, <span class="keywordtype">save</span> :: first_time = .true.</div><div class="line"><a name="l01311"></a><span class="lineno"> 1311</span>&#160;      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_ddpdd)&#39;</span></div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span>&#160;</div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span>&#160;</div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span>&#160;      <span class="keyword">call </span>shr_reprosumx86_fix_start (old_cw)</div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span>&#160;</div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>&#160;      <span class="keywordflow">if</span> (first_time) <span class="keywordflow">then</span></div><div class="line"><a name="l01318"></a><span class="lineno"> 1318</span>&#160;         <span class="keyword">call </span>mpi_op_create(<a class="code" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">ddpdd</a>, .true., mpi_sumdd, ierr)</div><div class="line"><a name="l01319"></a><span class="lineno"> 1319</span>&#160;         first_time = .false.</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span>&#160;<span class="keywordflow">      endif</span></div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span>&#160;</div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>&#160;      <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span>&#160;         arr_lsum_dd(ifld) = (0.0_r8,0.0_r8)</div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span>&#160;</div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>&#160;         <span class="keywordflow">do</span> isum=1,nsummands</div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span>&#160;                        </div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>&#160;            <span class="comment">! Compute arr(isum,ifld) + arr_lsum_dd(ifld) using Knuth&#39;&#39;s </span></div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>&#160;            <span class="comment">! trick.</span></div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span>&#160;            t1 = arr(isum,ifld) + <span class="keywordtype">real</span>(arr_lsum_dd(ifld))</div><div class="line"><a name="l01330"></a><span class="lineno"> 1330</span>&#160;            e  = t1 - arr(isum,ifld)</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>&#160;            t2 = ((<span class="keywordtype">real(arr_lsum_dd(ifld))</span> - e) &amp;</div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span>&#160;                  + (arr(isum,ifld) - (t1 - e))) &amp;</div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>&#160;                 + aimag(arr_lsum_dd(ifld))</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span>&#160;</div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>&#160;            <span class="comment">! The result is t1 + t2, after normalization.</span></div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>&#160;            arr_lsum_dd(ifld) = cmplx( t1 + t2, t2 - ((t1 + t2) - t1), r8 )</div><div class="line"><a name="l01337"></a><span class="lineno"> 1337</span>&#160;<span class="keywordflow">         enddo</span></div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span>&#160;</div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l01340"></a><span class="lineno"> 1340</span>&#160;</div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span>&#160;      <span class="keyword">call </span>mpi_allreduce (arr_lsum_dd, arr_gsum_dd, nflds, &amp;</div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span>&#160;                          mpi_complex16, mpi_sumdd, mpi_comm, ierr)</div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span>&#160;      <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span>&#160;         arr_gsum(ifld) = <span class="keywordtype">real</span>(arr_gsum_dd(ifld))</div><div class="line"><a name="l01345"></a><span class="lineno"> 1345</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</span>&#160;</div><div class="line"><a name="l01347"></a><span class="lineno"> 1347</span>&#160;      <span class="keyword">call </span>shr_reprosumx86_fix_end (old_cw)</div><div class="line"><a name="l01348"></a><span class="lineno"> 1348</span>&#160;</div><div class="line"><a name="l01349"></a><span class="lineno"> 1349</span>&#160;<span class="keyword">   end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">oasis_reprosum_ddpdd</a></div><div class="line"><a name="l01350"></a><span class="lineno"> 1350</span>&#160;</div><div class="line"><a name="l01351"></a><span class="lineno"> 1351</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01352"></a><span class="lineno"> 1352</span>&#160;</div><div class="line"><a name="l01353"></a><span class="lineno"> 1353</span>&#160;<span class="keyword">   subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">ddpdd</a> (dda, ddb, len, itype)</div><div class="line"><a name="l01354"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690"> 1354</a></span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01355"></a><span class="lineno"> 1355</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01356"></a><span class="lineno"> 1356</span>&#160;<span class="comment">! Purpose: </span></div><div class="line"><a name="l01357"></a><span class="lineno"> 1357</span>&#160;<span class="comment">! Modification of original codes written by David H. Bailey    </span></div><div class="line"><a name="l01358"></a><span class="lineno"> 1358</span>&#160;<span class="comment">! This subroutine computes ddb(i) = dda(i)+ddb(i)</span></div><div class="line"><a name="l01359"></a><span class="lineno"> 1359</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01360"></a><span class="lineno"> 1360</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01361"></a><span class="lineno"> 1361</span>&#160;</div><div class="line"><a name="l01362"></a><span class="lineno"> 1362</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l01363"></a><span class="lineno"> 1363</span>&#160;</div><div class="line"><a name="l01364"></a><span class="lineno"> 1364</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span>        :: len       <span class="comment">! array length</span></div><div class="line"><a name="l01365"></a><span class="lineno"> 1365</span>&#160;      <span class="keywordtype">complex(r8)</span>, <span class="keywordtype">intent(in)</span>    :: dda(len)  <span class="comment">! input</span></div><div class="line"><a name="l01366"></a><span class="lineno"> 1366</span>&#160;      <span class="keywordtype">complex(r8)</span>, <span class="keywordtype">intent(inout)</span> :: ddb(len)  <span class="comment">! result</span></div><div class="line"><a name="l01367"></a><span class="lineno"> 1367</span>&#160;      <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span>        :: itype     <span class="comment">! unused</span></div><div class="line"><a name="l01368"></a><span class="lineno"> 1368</span>&#160;</div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span>&#160;</div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span>&#160;      <span class="keywordtype">real(r8)</span> e, t1, t2</div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span>&#160;      <span class="keywordtype">integer</span> i</div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span>&#160;      <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_mod:DDPDD)&#39;</span></div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span>&#160;</div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span>&#160;</div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span>&#160;      <span class="keywordflow">do</span> i = 1, len</div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span>&#160;<span class="comment">!   Compute dda + ddb using Knuth&#39;s trick.</span></div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>&#160;         t1 = <span class="keywordtype">real(dda(i))</span> + <span class="keywordtype">real</span>(ddb(i))</div><div class="line"><a name="l01380"></a><span class="lineno"> 1380</span>&#160;         e  = t1 - <span class="keywordtype">real</span>(dda(i))</div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>&#160;         t2 = ((<span class="keywordtype">real(ddb(i))</span> - e) + (<span class="keywordtype">real(dda(i))</span> - (t1 - e))) &amp;</div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span>&#160;              + aimag(dda(i)) + aimag(ddb(i))</div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span>&#160;</div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span>&#160;<span class="comment">!   The result is t1 + t2, after normalization.</span></div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span>&#160;         ddb(i) = cmplx( t1 + t2, t2 - ((t1 + t2) - t1), r8 )</div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span>&#160;<span class="keywordflow">      enddo</span></div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span>&#160;</div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span>&#160;</div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</span>&#160;<span class="keyword">   end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">ddpdd</a></div><div class="line"><a name="l01390"></a><span class="lineno"> 1390</span>&#160;</div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span>&#160;</div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span>&#160;<span class="keyword">  subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">split_indices</a>(total,num_pieces,ibeg,iend)</div><div class="line"><a name="l01394"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004"> 1394</a></span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span>&#160;<span class="comment">! Purpose: </span></div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span>&#160;<span class="comment">! Split range into &#39;num_pieces&#39;</span></div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span>&#160;<span class="comment">! </span></div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span>&#160;<span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span>&#160;</div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span>&#160;<span class="comment">! Arguments</span></div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span>&#160;</div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span>&#160;    <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: total</div><div class="line"><a name="l01404"></a><span class="lineno"> 1404</span>&#160;    <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: num_pieces</div><div class="line"><a name="l01405"></a><span class="lineno"> 1405</span>&#160;    <span class="keywordtype">integer</span>, <span class="keywordtype">intent(out)</span> :: ibeg(num_pieces), iend(num_pieces)</div><div class="line"><a name="l01406"></a><span class="lineno"> 1406</span>&#160;</div><div class="line"><a name="l01407"></a><span class="lineno"> 1407</span>&#160;<span class="comment">! Local workspace</span></div><div class="line"><a name="l01408"></a><span class="lineno"> 1408</span>&#160;</div><div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>&#160;    <span class="keywordtype">integer</span> :: itmp1, itmp2, ioffset, i</div><div class="line"><a name="l01410"></a><span class="lineno"> 1410</span>&#160;    <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">&#39;(oasis_reprosum_mod:split_indices)&#39;</span></div><div class="line"><a name="l01411"></a><span class="lineno"> 1411</span>&#160;</div><div class="line"><a name="l01412"></a><span class="lineno"> 1412</span>&#160;<span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01413"></a><span class="lineno"> 1413</span>&#160;</div><div class="line"><a name="l01414"></a><span class="lineno"> 1414</span>&#160;    itmp1 = total/num_pieces</div><div class="line"><a name="l01415"></a><span class="lineno"> 1415</span>&#160;    itmp2 = mod(total,num_pieces)</div><div class="line"><a name="l01416"></a><span class="lineno"> 1416</span>&#160;    ioffset = 0</div><div class="line"><a name="l01417"></a><span class="lineno"> 1417</span>&#160;    <span class="keywordflow">do</span> i=1,itmp2</div><div class="line"><a name="l01418"></a><span class="lineno"> 1418</span>&#160;       ibeg(i) = ioffset + 1</div><div class="line"><a name="l01419"></a><span class="lineno"> 1419</span>&#160;       iend(i) = ioffset + (itmp1+1)</div><div class="line"><a name="l01420"></a><span class="lineno"> 1420</span>&#160;       ioffset = iend(i)</div><div class="line"><a name="l01421"></a><span class="lineno"> 1421</span>&#160;<span class="keywordflow">    enddo</span></div><div class="line"><a name="l01422"></a><span class="lineno"> 1422</span>&#160;    <span class="keywordflow">do</span> i=itmp2+1,num_pieces</div><div class="line"><a name="l01423"></a><span class="lineno"> 1423</span>&#160;       ibeg(i) = ioffset + 1</div><div class="line"><a name="l01424"></a><span class="lineno"> 1424</span>&#160;       <span class="keywordflow">if</span> (ibeg(i) &gt; total) <span class="keywordflow">then</span></div><div class="line"><a name="l01425"></a><span class="lineno"> 1425</span>&#160;          iend(i) = ibeg(i) - 1</div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span>&#160;       <span class="keywordflow">else</span></div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span>&#160;          iend(i) = ioffset + itmp1</div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span>&#160;          ioffset = iend(i)</div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span>&#160;<span class="keywordflow">       endif</span></div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span>&#160;<span class="keywordflow">    enddo</span></div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span>&#160;</div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span>&#160;<span class="keyword">  end subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">split_indices</a></div><div class="line"><a name="l01433"></a><span class="lineno"> 1433</span>&#160;</div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span>&#160;<span class="comment">!========================================================================</span></div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span>&#160;</div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span>&#160;<span class="keyword">end module </span><a class="code" href="namespacemod__oasis__reprosum.html">mod_oasis_reprosum</a></div><div class="ttc" id="namespacemod__oasis__sys_html"><div class="ttname"><a href="namespacemod__oasis__sys.html">mod_oasis_sys</a></div><div class="ttdoc">System type methods. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__sys_8_f90_source.html#l00004">mod_oasis_sys.F90:4</a></div></div>
70<div class="ttc" id="namespacemod__oasis__reprosum_html_a7ea3c5bad9db10c80eb528453602aed3"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">mod_oasis_reprosum::oasis_reprosum_int</a></div><div class="ttdeci">subroutine oasis_reprosum_int(arr, arr_gsum, nsummands, dsummands, nflds, arr_max_shift, arr_gmax_exp, max_levels, max_level, validate, recompute, omp_nthreads, mpi_comm)</div><div class="ttdoc">Compute the global sum of each field in &quot;arr&quot; using the indicated communicator with a reproducible ye...</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00751">mod_oasis_reprosum.F90:751</a></div></div>
71<div class="ttc" id="namespacemod__oasis__data_html"><div class="ttname"><a href="namespacemod__oasis__data.html">mod_oasis_data</a></div><div class="ttdoc">Provides a common location for several OASIS variables. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__data_8_f90_source.html#l00004">mod_oasis_data.F90:4</a></div></div>
72<div class="ttc" id="namespacemod__oasis__reprosum_html_a398280c78cff126b7622dbd96af675e4"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">mod_oasis_reprosum::oasis_reprosum_calc</a></div><div class="ttdeci">subroutine, public oasis_reprosum_calc(arr, arr_gsum, nsummands, dsummands, nflds, ddpdd_sum, arr_gbl_max, arr_gbl_max_out, arr_max_levels, arr_max_levels_out, gbl_max_nsummands, gbl_max_nsummands_out, gbl_count, repro_sum_validate, repro_sum_stats, rel_diff, commid)</div><div class="ttdoc">Compute the global sum of each field in &quot;arr&quot; using the indicated communicator with a reproducible ye...</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00267">mod_oasis_reprosum.F90:267</a></div></div>
73<div class="ttc" id="namespacemod__oasis__reprosum_html_a48a084c213295b9eaa78c0552a10741a"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a48a084c213295b9eaa78c0552a10741a">mod_oasis_reprosum::detailed_timing</a></div><div class="ttdeci">logical detailed_timing</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00089">mod_oasis_reprosum.F90:89</a></div></div>
74<div class="ttc" id="namespacemod__oasis__data_html_af0f17b70f48598dac24f2e41ca0da4b2"><div class="ttname"><a href="namespacemod__oasis__data.html#af0f17b70f48598dac24f2e41ca0da4b2">mod_oasis_data::nulprt</a></div><div class="ttdeci">integer(kind=ip_intwp_p) nulprt</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__data_8_f90_source.html#l00020">mod_oasis_data.F90:20</a></div></div>
75<div class="ttc" id="namespacemod__oasis__reprosum_html_ad6b4914497a4961339a1c95bfb807660"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">mod_oasis_reprosum::oasis_reprosum_setopts</a></div><div class="ttdeci">subroutine, public oasis_reprosum_setopts(repro_sum_use_ddpdd_in, repro_sum_rel_diff_max_in, repro_sum_recompute_in, repro_sum_master, repro_sum_logunit)</div><div class="ttdoc">Set runtime options. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00105">mod_oasis_reprosum.F90:105</a></div></div>
76<div class="ttc" id="namespacemod__oasis__sys_html_af1943a0ce0c5765ba8a714379065464f"><div class="ttname"><a href="namespacemod__oasis__sys.html#af1943a0ce0c5765ba8a714379065464f">mod_oasis_sys::oasis_abort</a></div><div class="ttdeci">subroutine, public oasis_abort(id_compid, cd_routine, cd_message, file, line, rcode)</div><div class="ttdoc">OASIS abort method, publically available to users. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__sys_8_f90_source.html#l00044">mod_oasis_sys.F90:44</a></div></div>
77<div class="ttc" id="namespacemod__oasis__kinds_html_a1e3c90ad6857ff8675d3e2548d516200"><div class="ttname"><a href="namespacemod__oasis__kinds.html#a1e3c90ad6857ff8675d3e2548d516200">mod_oasis_kinds::ip_i4_p</a></div><div class="ttdeci">integer, parameter ip_i4_p</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__kinds_8_f90_source.html#l00036">mod_oasis_kinds.F90:36</a></div></div>
78<div class="ttc" id="namespacemod__oasis__kinds_html_a6129e979d32b55cb020fb76068ce9808"><div class="ttname"><a href="namespacemod__oasis__kinds.html#a6129e979d32b55cb020fb76068ce9808">mod_oasis_kinds::ip_r8_p</a></div><div class="ttdeci">integer, parameter ip_r8_p</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__kinds_8_f90_source.html#l00043">mod_oasis_kinds.F90:43</a></div></div>
79<div class="ttc" id="namespacemod__oasis__reprosum_html_a24f9efa3ca739811888206eaf71d4690"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">mod_oasis_reprosum::ddpdd</a></div><div class="ttdeci">subroutine, private ddpdd(dda, ddb, len, itype)</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l01354">mod_oasis_reprosum.F90:1354</a></div></div>
80<div class="ttc" id="namespacemod__oasis__kinds_html"><div class="ttname"><a href="namespacemod__oasis__kinds.html">mod_oasis_kinds</a></div><div class="ttdoc">Defines kinds for OASIS. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__kinds_8_f90_source.html#l00007">mod_oasis_kinds.F90:7</a></div></div>
81<div class="ttc" id="namespacemod__oasis__reprosum_html_a6a80f4226d35d0842617e6eb6c9406e1"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">mod_oasis_reprosum::oasis_reprosum_reldiffmax</a></div><div class="ttdeci">real(r8), public oasis_reprosum_reldiffmax</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00075">mod_oasis_reprosum.F90:75</a></div></div>
82<div class="ttc" id="namespacemod__oasis__mpi_html"><div class="ttname"><a href="namespacemod__oasis__mpi.html">mod_oasis_mpi</a></div><div class="ttdoc">Provides a generic and simpler interface into MPI calls for OASIS. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__mpi_8_f90_source.html#l00003">mod_oasis_mpi.F90:3</a></div></div>
83<div class="ttc" id="namespacemod__oasis__reprosum_html_a0f773341a7a6883b2f65d395a6b25b08"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a0f773341a7a6883b2f65d395a6b25b08">mod_oasis_reprosum::oasis_reprosum_tolexceeded</a></div><div class="ttdeci">logical function, public oasis_reprosum_tolexceeded(name, nflds, master, logunit, rel_diff)</div><div class="ttdoc">Test whether distributed sum exceeds tolerance and print out a warning message. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l01195">mod_oasis_reprosum.F90:1195</a></div></div>
84<div class="ttc" id="namespacemod__oasis__timer_html_a13a16c07778911b45f3f42a542348019"><div class="ttname"><a href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">mod_oasis_timer::oasis_timer_start</a></div><div class="ttdeci">subroutine, public oasis_timer_start(timer_label, barrier)</div><div class="ttdoc">Start a timer. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__timer_8_f90_source.html#l00134">mod_oasis_timer.F90:134</a></div></div>
85<div class="ttc" id="namespacemod__oasis__timer_html_a01a76142a72cc6bc845736e79ad99bd3"><div class="ttname"><a href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">mod_oasis_timer::oasis_timer_stop</a></div><div class="ttdeci">subroutine, public oasis_timer_stop(timer_label)</div><div class="ttdoc">Stop a timer. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__timer_8_f90_source.html#l00178">mod_oasis_timer.F90:178</a></div></div>
86<div class="ttc" id="namespacemod__oasis__sys_html_a0f6e0376585c6f39d653f5ab05e80e8f"><div class="ttname"><a href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">mod_oasis_sys::estr</a></div><div class="ttdeci">character(len= *), parameter, public estr</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__sys_8_f90_source.html#l00012">mod_oasis_sys.F90:12</a></div></div>
87<div class="ttc" id="namespacemod__oasis__mpi_html_aaf477847380fbbbc1341520b0b3e66da"><div class="ttname"><a href="namespacemod__oasis__mpi.html#aaf477847380fbbbc1341520b0b3e66da">mod_oasis_mpi::oasis_mpi_barrier</a></div><div class="ttdeci">subroutine, public oasis_mpi_barrier(comm, string)</div><div class="ttdoc">Call MPI_BARRIER for a particular communicator. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__mpi_8_f90_source.html#l02697">mod_oasis_mpi.F90:2697</a></div></div>
88<div class="ttc" id="namespacemod__oasis__reprosum_html_aab06d0de9a82726d347ac52bb276355d"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d">mod_oasis_reprosum::repro_sum_use_ddpdd</a></div><div class="ttdeci">logical repro_sum_use_ddpdd</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00088">mod_oasis_reprosum.F90:88</a></div></div>
89<div class="ttc" id="namespacemod__oasis__reprosum_html_affb09b4ec9b001bf2ebbc51feebd8004"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#affb09b4ec9b001bf2ebbc51feebd8004">mod_oasis_reprosum::split_indices</a></div><div class="ttdeci">subroutine, private split_indices(total, num_pieces, ibeg, iend)</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l01394">mod_oasis_reprosum.F90:1394</a></div></div>
90<div class="ttc" id="namespacemod__oasis__reprosum_html_afe237975bf8e718f32523c3929911102"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">mod_oasis_reprosum::oasis_reprosum_ddpdd</a></div><div class="ttdeci">subroutine oasis_reprosum_ddpdd(arr, arr_gsum, nsummands, dsummands, nflds, mpi_comm)</div><div class="ttdoc">Compute the global sum of each field in &quot;arr&quot; using the indicated communicator with a reproducible ye...</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l01281">mod_oasis_reprosum.F90:1281</a></div></div>
91<div class="ttc" id="namespacemod__oasis__kinds_html_a8dbd6aa926433f02df94bca462f122df"><div class="ttname"><a href="namespacemod__oasis__kinds.html#a8dbd6aa926433f02df94bca462f122df">mod_oasis_kinds::ip_i8_p</a></div><div class="ttdeci">integer, parameter ip_i8_p</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__kinds_8_f90_source.html#l00040">mod_oasis_kinds.F90:40</a></div></div>
92<div class="ttc" id="namespacemod__oasis__timer_html"><div class="ttname"><a href="namespacemod__oasis__timer.html">mod_oasis_timer</a></div><div class="ttdoc">Performance timer methods. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__timer_8_f90_source.html#l00039">mod_oasis_timer.F90:39</a></div></div>
93<div class="ttc" id="namespacemod__oasis__reprosum_html"><div class="ttname"><a href="namespacemod__oasis__reprosum.html">mod_oasis_reprosum</a></div><div class="ttdoc">OASIS reproducible sum method from P. Worley. </div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00004">mod_oasis_reprosum.F90:4</a></div></div>
94<div class="ttc" id="namespacemod__oasis__reprosum_html_a89140c250e34a903bc9aa546574988d6"><div class="ttname"><a href="namespacemod__oasis__reprosum.html#a89140c250e34a903bc9aa546574988d6">mod_oasis_reprosum::oasis_reprosum_recompute</a></div><div class="ttdeci">logical, public oasis_reprosum_recompute</div><div class="ttdef"><b>Definition:</b> <a href="mod__oasis__reprosum_8_f90_source.html#l00073">mod_oasis_reprosum.F90:73</a></div></div>
95</div><!-- fragment --></div><!-- contents -->
96<!-- start footer part -->
97<hr class="footer"/><address class="footer"><small>
98Generated on Fri Jun 29 2018 20:52:28 for Oasis3-MCT by &#160;<a href="http://www.doxygen.org/index.html">
99<img class="footer" src="doxygen.png" alt="doxygen"/>
100</a> 1.8.13
101</small></address>
102</body>
103</html>
Note: See TracBrowser for help on using the repository browser.