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> <span class="comment"></span></div><div class="line"><a name="l00002"></a><span class="lineno"> 2</span> <span class="comment">!> OASIS reproducible sum method from P. Worley</span></div><div class="line"><a name="l00003"></a><span class="lineno"> 3</span> </div><div class="line"><a name="l00004"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html"> 4</a></span> <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> </div><div class="line"><a name="l00006"></a><span class="lineno"> 6</span> <span class="comment">!----------------------------------------------------------------------- </span></div><div class="line"><a name="l00007"></a><span class="lineno"> 7</span> <span class="comment">! </span></div><div class="line"><a name="l00008"></a><span class="lineno"> 8</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00009"></a><span class="lineno"> 9</span> <span class="comment">!> 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> <span class="comment">!> subcommunicator</span></div><div class="line"><a name="l00011"></a><span class="lineno"> 11</span> <span class="comment">!</span></div><div class="line"><a name="l00012"></a><span class="lineno"> 12</span> <span class="comment">! Methods: </span><span class="comment"></span></div><div class="line"><a name="l00013"></a><span class="lineno"> 13</span> <span class="comment">!> 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> <span class="comment">!> scalable, nonreproducible algorithm:</span></div><div class="line"><a name="l00015"></a><span class="lineno"> 15</span> <span class="comment">!> * Reproducible (scalable): </span></div><div class="line"><a name="l00016"></a><span class="lineno"> 16</span> <span class="comment">!> Convert to fixed point (integer vector representation) to enable</span></div><div class="line"><a name="l00017"></a><span class="lineno"> 17</span> <span class="comment">!> reproducibility when using MPI_Allreduce</span></div><div class="line"><a name="l00018"></a><span class="lineno"> 18</span> <span class="comment">!> * Alternative usually reproducible (scalable):</span></div><div class="line"><a name="l00019"></a><span class="lineno"> 19</span> <span class="comment">!> Use parallel double-double algorithm due to Helen He and </span></div><div class="line"><a name="l00020"></a><span class="lineno"> 20</span> <span class="comment">!> Chris Ding, based on David Bailey's/Don Knuth's DDPDD algorithm</span></div><div class="line"><a name="l00021"></a><span class="lineno"> 21</span> <span class="comment">!> * Nonreproducible (scalable): </span></div><div class="line"><a name="l00022"></a><span class="lineno"> 22</span> <span class="comment">!> Floating point and MPI_Allreduce based.</span></div><div class="line"><a name="l00023"></a><span class="lineno"> 23</span> <span class="comment">!> If computing both reproducible and nonreproducible sums, compare </span></div><div class="line"><a name="l00024"></a><span class="lineno"> 24</span> <span class="comment">!> these and report relative difference (if absolute difference</span></div><div class="line"><a name="l00025"></a><span class="lineno"> 25</span> <span class="comment">!> less than sum) or absolute difference back to calling routine.</span></div><div class="line"><a name="l00026"></a><span class="lineno"> 26</span> <span class="comment">!</span></div><div class="line"><a name="l00027"></a><span class="lineno"> 27</span> <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> <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> <span class="comment">! algorithm)</span></div><div class="line"><a name="l00030"></a><span class="lineno"> 30</span> <span class="comment">!</span></div><div class="line"><a name="l00031"></a><span class="lineno"> 31</span> <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> <span class="comment">! CESM1.2.</span></div><div class="line"><a name="l00033"></a><span class="lineno"> 33</span> <span class="comment">! </span></div><div class="line"><a name="l00034"></a><span class="lineno"> 34</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00035"></a><span class="lineno"> 35</span> </div><div class="line"><a name="l00036"></a><span class="lineno"> 36</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00037"></a><span class="lineno"> 37</span> <span class="comment">!- use statements ------------------------------------------------------</span></div><div class="line"><a name="l00038"></a><span class="lineno"> 38</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00039"></a><span class="lineno"> 39</span> <span class="preprocessor">#if ( defined noI8 )</span></div><div class="line"><a name="l00040"></a><span class="lineno"> 40</span> <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>  <span class="keywordtype">use </span><a class="code" href="namespacemod__oasis__kinds.html">mod_oasis_kinds</a><span class="keywordtype">, only</span>: r8 => <a class="code" href="namespacemod__oasis__kinds.html#a6129e979d32b55cb020fb76068ce9808">ip_r8_p</a>, i8 => <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> <span class="preprocessor">#else</span></div><div class="line"><a name="l00043"></a><span class="lineno"> 43</span> <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 => <a class="code" href="namespacemod__oasis__kinds.html#a6129e979d32b55cb020fb76068ce9808">ip_r8_p</a>, i8 => <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> <span class="preprocessor">#endif</span></div><div class="line"><a name="l00045"></a><span class="lineno"> 45</span> <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>  <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>  <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>  <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> </div><div class="line"><a name="l00050"></a><span class="lineno"> 50</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00051"></a><span class="lineno"> 51</span> <span class="comment">!- module boilerplate --------------------------------------------------</span></div><div class="line"><a name="l00052"></a><span class="lineno"> 52</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00053"></a><span class="lineno"> 53</span>  <span class="keywordtype">implicit none</span></div><div class="line"><a name="l00054"></a><span class="lineno"> 54</span>  <span class="keywordtype">private</span></div><div class="line"><a name="l00055"></a><span class="lineno"> 55</span> </div><div class="line"><a name="l00056"></a><span class="lineno"> 56</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00057"></a><span class="lineno"> 57</span> <span class="comment">!- include statements --------------------------------------------------</span></div><div class="line"><a name="l00058"></a><span class="lineno"> 58</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00059"></a><span class="lineno"> 59</span> <span class="preprocessor">#include <mpif.h></span></div><div class="line"><a name="l00060"></a><span class="lineno"> 60</span> <span class="preprocessor"></span></div><div class="line"><a name="l00061"></a><span class="lineno"> 61</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00062"></a><span class="lineno"> 62</span> <span class="comment">! Public interfaces ----------------------------------------------------</span></div><div class="line"><a name="l00063"></a><span class="lineno"> 63</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00064"></a><span class="lineno"> 64</span>  <span class="keywordtype">public</span> :: &</div><div class="line"><a name="l00065"></a><span class="lineno"> 65</span>  <a class="code" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">oasis_reprosum_setopts</a>, &<span class="comment">! set runtime options</span></div><div class="line"><a name="l00066"></a><span class="lineno"> 66</span>  <a class="code" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">oasis_reprosum_calc</a>, &<span class="comment">! calculate distributed sum</span></div><div class="line"><a name="l00067"></a><span class="lineno"> 67</span>  <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>  <span class="comment">! differences against the tolerance</span></div><div class="line"><a name="l00069"></a><span class="lineno"> 69</span> </div><div class="line"><a name="l00070"></a><span class="lineno"> 70</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00071"></a><span class="lineno"> 71</span> <span class="comment">! Public data ----------------------------------------------------------</span></div><div class="line"><a name="l00072"></a><span class="lineno"> 72</span> <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>  <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> </div><div class="line"><a name="l00075"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1"> 75</a></span>  <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> </div><div class="line"><a name="l00077"></a><span class="lineno"> 77</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00078"></a><span class="lineno"> 78</span> <span class="comment">! Private interfaces ---------------------------------------------------</span></div><div class="line"><a name="l00079"></a><span class="lineno"> 79</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00080"></a><span class="lineno"> 80</span>  <span class="keywordtype">private</span> :: &</div><div class="line"><a name="l00081"></a><span class="lineno"> 81</span>  <a class="code" href="namespacemod__oasis__reprosum.html#a24f9efa3ca739811888206eaf71d4690">ddpdd</a>, &<span class="comment">! double-double sum routine</span></div><div class="line"><a name="l00082"></a><span class="lineno"> 82</span>  <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> </div><div class="line"><a name="l00084"></a><span class="lineno"> 84</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00085"></a><span class="lineno"> 85</span> <span class="comment">! Private data ---------------------------------------------------------</span></div><div class="line"><a name="l00086"></a><span class="lineno"> 86</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00087"></a><span class="lineno"> 87</span> </div><div class="line"><a name="l00088"></a><span class="lineno"><a class="line" href="namespacemod__oasis__reprosum.html#aab06d0de9a82726d347ac52bb276355d"> 88</a></span>  <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>  <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> </div><div class="line"><a name="l00091"></a><span class="lineno"> 91</span>  <span class="keyword">CONTAINS</span></div><div class="line"><a name="l00092"></a><span class="lineno"> 92</span> </div><div class="line"><a name="l00093"></a><span class="lineno"> 93</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l00094"></a><span class="lineno"> 94</span> <span class="comment">!----------------------------------------------------------------------- </span></div><div class="line"><a name="l00095"></a><span class="lineno"> 95</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00096"></a><span class="lineno"> 96</span> <span class="comment">!> Set runtime options</span></div><div class="line"><a name="l00097"></a><span class="lineno"> 97</span> <span class="comment">! Author: P. Worley</span></div><div class="line"><a name="l00098"></a><span class="lineno"> 98</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00099"></a><span class="lineno"> 99</span> </div><div class="line"><a name="l00100"></a><span class="lineno"> 100</span> <span class="keyword"> subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#ad6b4914497a4961339a1c95bfb807660">oasis_reprosum_setopts</a>(repro_sum_use_ddpdd_in, &</div><div class="line"><a name="l00101"></a><span class="lineno"> 101</span>  repro_sum_rel_diff_max_in, &</div><div class="line"><a name="l00102"></a><span class="lineno"> 102</span>  repro_sum_recompute_in, &</div><div class="line"><a name="l00103"></a><span class="lineno"> 103</span>  repro_sum_master, &</div><div class="line"><a name="l00104"></a><span class="lineno"> 104</span>  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> </div><div class="line"><a name="l00106"></a><span class="lineno"> 106</span> <span class="comment">!------------------------------Arguments--------------------------------</span></div><div class="line"><a name="l00107"></a><span class="lineno"> 107</span>  <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>  <span class="comment">!< Use DDPDD algorithm instead of fixed precision algorithm</span></div><div class="line"><a name="l00109"></a><span class="lineno"> 109</span>  <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>  <span class="comment">!< maximum permissible difference between reproducible and </span></div><div class="line"><a name="l00111"></a><span class="lineno"> 111</span> <span class="comment"> !< nonreproducible sums</span></div><div class="line"><a name="l00112"></a><span class="lineno"> 112</span>  <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>  <span class="comment">!< recompute using different algorithm when difference between </span></div><div class="line"><a name="l00114"></a><span class="lineno"> 114</span> <span class="comment"> !< reproducible and nonreproducible sums is too great </span></div><div class="line"><a name="l00115"></a><span class="lineno"> 115</span>  <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>  <span class="comment">!< flag indicating whether this process should output</span></div><div class="line"><a name="l00117"></a><span class="lineno"> 117</span> <span class="comment"> !< log messages</span></div><div class="line"><a name="l00118"></a><span class="lineno"> 118</span>  <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>  <span class="comment">!< unit number for log messages</span></div><div class="line"><a name="l00120"></a><span class="lineno"> 120</span> <span class="comment">!---------------------------Local Workspace-----------------------------</span></div><div class="line"><a name="l00121"></a><span class="lineno"> 121</span>  <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>  <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>  <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>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_setopts)'</span></div><div class="line"><a name="l00125"></a><span class="lineno"> 125</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00126"></a><span class="lineno"> 126</span> </div><div class="line"><a name="l00127"></a><span class="lineno"> 127</span>  <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>  master = repro_sum_master</div><div class="line"><a name="l00129"></a><span class="lineno"> 129</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00130"></a><span class="lineno"> 130</span>  master = .false.</div><div class="line"><a name="l00131"></a><span class="lineno"> 131</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00132"></a><span class="lineno"> 132</span> </div><div class="line"><a name="l00133"></a><span class="lineno"> 133</span>  <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>  llogunit = repro_sum_logunit</div><div class="line"><a name="l00135"></a><span class="lineno"> 135</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00136"></a><span class="lineno"> 136</span>  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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00138"></a><span class="lineno"> 138</span> </div><div class="line"><a name="l00139"></a><span class="lineno"> 139</span>  <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>  <span class="keyword">write</span>(llogunit,*) subname,<a class="code" href="namespacemod__oasis__sys.html#a0f6e0376585c6f39d653f5ab05e80e8f">estr</a>,<span class="stringliteral">' can only be called once'</span></div><div class="line"><a name="l00141"></a><span class="lineno"> 141</span>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00143"></a><span class="lineno"> 143</span>  firstcall = .false.</div><div class="line"><a name="l00144"></a><span class="lineno"> 144</span> </div><div class="line"><a name="l00145"></a><span class="lineno"> 145</span>  <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>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00148"></a><span class="lineno"> 148</span>  <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>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00151"></a><span class="lineno"> 151</span>  <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>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00154"></a><span class="lineno"> 154</span>  <span class="keywordflow">if</span> (master) <span class="keywordflow">then</span></div><div class="line"><a name="l00155"></a><span class="lineno"> 155</span>  <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>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00157"></a><span class="lineno"> 157</span>  <span class="stringliteral">'Using double-double-based (scalable) usually reproducible '</span>, &</div><div class="line"><a name="l00158"></a><span class="lineno"> 158</span>  <span class="stringliteral">'distributed sum algorithm'</span></div><div class="line"><a name="l00159"></a><span class="lineno"> 159</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00160"></a><span class="lineno"> 160</span>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00161"></a><span class="lineno"> 161</span>  <span class="stringliteral">'Using fixed-point-based (scalable) reproducible '</span>, &</div><div class="line"><a name="l00162"></a><span class="lineno"> 162</span>  <span class="stringliteral">'distributed sum algorithm'</span></div><div class="line"><a name="l00163"></a><span class="lineno"> 163</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00164"></a><span class="lineno"> 164</span> </div><div class="line"><a name="l00165"></a><span class="lineno"> 165</span>  <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> >= 0._r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00166"></a><span class="lineno"> 166</span>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00167"></a><span class="lineno"> 167</span>  <span class="stringliteral">' with a maximum relative error tolerance of '</span>, &</div><div class="line"><a name="l00168"></a><span class="lineno"> 168</span>  <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>  <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>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00171"></a><span class="lineno"> 171</span>  <span class="stringliteral">'If tolerance exceeded, sum is recomputed using '</span>, &</div><div class="line"><a name="l00172"></a><span class="lineno"> 172</span>  <span class="stringliteral">'a serial algorithm.'</span></div><div class="line"><a name="l00173"></a><span class="lineno"> 173</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00174"></a><span class="lineno"> 174</span>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00175"></a><span class="lineno"> 175</span>  <span class="stringliteral">'If tolerance exceeded, fixed-precision is sum used '</span>, &</div><div class="line"><a name="l00176"></a><span class="lineno"> 176</span>  <span class="stringliteral">'but a warning is output.'</span></div><div class="line"><a name="l00177"></a><span class="lineno"> 177</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00178"></a><span class="lineno"> 178</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00179"></a><span class="lineno"> 179</span>  <span class="keyword">write</span>(llogunit,*) subname, &</div><div class="line"><a name="l00180"></a><span class="lineno"> 180</span>  <span class="stringliteral">'and not comparing with floating point algorithms.'</span></div><div class="line"><a name="l00181"></a><span class="lineno"> 181</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00182"></a><span class="lineno"> 182</span> </div><div class="line"><a name="l00183"></a><span class="lineno"> 183</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00184"></a><span class="lineno"> 184</span> <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> </div><div class="line"><a name="l00186"></a><span class="lineno"> 186</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l00187"></a><span class="lineno"> 187</span> </div><div class="line"><a name="l00188"></a><span class="lineno"> 188</span> <span class="comment">! </span></div><div class="line"><a name="l00189"></a><span class="lineno"> 189</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00190"></a><span class="lineno"> 190</span> <span class="comment">!> Compute the global sum of each field in "arr" using the indicated </span></div><div class="line"><a name="l00191"></a><span class="lineno"> 191</span> <span class="comment">!> communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l00192"></a><span class="lineno"> 192</span> <span class="comment">!> on a fixed point algorithm. An alternative is to use an "almost</span></div><div class="line"><a name="l00193"></a><span class="lineno"> 193</span> <span class="comment">!> always reproducible" floating point algorithm.</span></div><div class="line"><a name="l00194"></a><span class="lineno"> 194</span> <span class="comment">!</span></div><div class="line"><a name="l00195"></a><span class="lineno"> 195</span> <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> <span class="comment">! number of "levels" of integer expansion. The algorithm will calculate</span></div><div class="line"><a name="l00197"></a><span class="lineno"> 197</span> <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> <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> <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> <span class="comment">! used to return the values used.</span></div><div class="line"><a name="l00201"></a><span class="lineno"> 201</span> <span class="comment">!</span></div><div class="line"><a name="l00202"></a><span class="lineno"> 202</span> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <span class="comment">! used to return the values used.</span></div><div class="line"><a name="l00214"></a><span class="lineno"> 214</span> <span class="comment">!</span></div><div class="line"><a name="l00215"></a><span class="lineno"> 215</span> <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> <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> <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> <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> <span class="comment">! it is > 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> <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> <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> <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> <span class="comment">! However, if set to < 1, the value will instead be calculated. If the </span></div><div class="line"><a name="l00224"></a><span class="lineno"> 224</span> <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> <span class="comment">! used (gbl_max_nsummands if >= 1; calculated otherwise) will be </span></div><div class="line"><a name="l00226"></a><span class="lineno"> 226</span> <span class="comment">! returned.</span></div><div class="line"><a name="l00227"></a><span class="lineno"> 227</span> <span class="comment">!</span></div><div class="line"><a name="l00228"></a><span class="lineno"> 228</span> <span class="comment">! If requested (by setting oasis_reprosum_reldiffmax >= 0.0 and passing in</span></div><div class="line"><a name="l00229"></a><span class="lineno"> 229</span> <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> <span class="comment">! nonreproducible floating point algorithm. </span></div><div class="line"><a name="l00231"></a><span class="lineno"> 231</span> <span class="comment">!</span></div><div class="line"><a name="l00232"></a><span class="lineno"> 232</span> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <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> <span class="comment">!</span></div><div class="line"><a name="l00246"></a><span class="lineno"> 246</span> <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> <span class="comment">! implementation of David Bailey's routine DDPDD by Helen He</span></div><div class="line"><a name="l00248"></a><span class="lineno"> 248</span> <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> <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> <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> <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> <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> <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> <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> <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> <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> <span class="comment">! hold. </span></div><div class="line"><a name="l00258"></a><span class="lineno"> 258</span> </div><div class="line"><a name="l00259"></a><span class="lineno"> 259</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00260"></a><span class="lineno"> 260</span> <span class="keyword"> subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a398280c78cff126b7622dbd96af675e4">oasis_reprosum_calc</a> (arr, arr_gsum, nsummands, dsummands, &</div><div class="line"><a name="l00261"></a><span class="lineno"> 261</span>  nflds, ddpdd_sum, &</div><div class="line"><a name="l00262"></a><span class="lineno"> 262</span>  arr_gbl_max, arr_gbl_max_out, &</div><div class="line"><a name="l00263"></a><span class="lineno"> 263</span>  arr_max_levels, arr_max_levels_out, &</div><div class="line"><a name="l00264"></a><span class="lineno"> 264</span>  gbl_max_nsummands, gbl_max_nsummands_out,&</div><div class="line"><a name="l00265"></a><span class="lineno"> 265</span>  gbl_count, repro_sum_validate, &</div><div class="line"><a name="l00266"></a><span class="lineno"> 266</span>  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> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00268"></a><span class="lineno"> 268</span> </div><div class="line"><a name="l00269"></a><span class="lineno"> 269</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l00270"></a><span class="lineno"> 270</span> </div><div class="line"><a name="l00271"></a><span class="lineno"> 271</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nsummands<span class="comment"> !< number of local summands</span></div><div class="line"><a name="l00272"></a><span class="lineno"> 272</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: dsummands<span class="comment"> !< declared first dimension</span></div><div class="line"><a name="l00273"></a><span class="lineno"> 273</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nflds<span class="comment"> !< number of fields</span></div><div class="line"><a name="l00274"></a><span class="lineno"> 274</span>  <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>  <span class="comment">!< input array</span></div><div class="line"><a name="l00276"></a><span class="lineno"> 276</span> </div><div class="line"><a name="l00277"></a><span class="lineno"> 277</span>  <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>  <span class="comment">!< global means</span></div><div class="line"><a name="l00279"></a><span class="lineno"> 279</span> </div><div class="line"><a name="l00280"></a><span class="lineno"> 280</span>  <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>  <span class="comment">!< use ddpdd algorithm instead</span></div><div class="line"><a name="l00282"></a><span class="lineno"> 282</span> <span class="comment"> !< of fixed precision algorithm</span></div><div class="line"><a name="l00283"></a><span class="lineno"> 283</span> </div><div class="line"><a name="l00284"></a><span class="lineno"> 284</span>  <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>  <span class="comment">!< upper bound on max(abs(arr))</span></div><div class="line"><a name="l00286"></a><span class="lineno"> 286</span> </div><div class="line"><a name="l00287"></a><span class="lineno"> 287</span>  <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>  <span class="comment">!< calculated upper bound on </span></div><div class="line"><a name="l00289"></a><span class="lineno"> 289</span> <span class="comment"> !< max(abs(arr))</span></div><div class="line"><a name="l00290"></a><span class="lineno"> 290</span> </div><div class="line"><a name="l00291"></a><span class="lineno"> 291</span>  <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>  <span class="comment">!< maximum number of levels of</span></div><div class="line"><a name="l00293"></a><span class="lineno"> 293</span> <span class="comment"> !< integer expansion to use</span></div><div class="line"><a name="l00294"></a><span class="lineno"> 294</span> </div><div class="line"><a name="l00295"></a><span class="lineno"> 295</span>  <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>  <span class="comment">!< output of number of levels of</span></div><div class="line"><a name="l00297"></a><span class="lineno"> 297</span> <span class="comment"> !< integer expansion to used</span></div><div class="line"><a name="l00298"></a><span class="lineno"> 298</span> </div><div class="line"><a name="l00299"></a><span class="lineno"> 299</span>  <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>  <span class="comment">!< maximum of nsummand over all</span></div><div class="line"><a name="l00301"></a><span class="lineno"> 301</span> <span class="comment"> !< processes</span></div><div class="line"><a name="l00302"></a><span class="lineno"> 302</span> </div><div class="line"><a name="l00303"></a><span class="lineno"> 303</span>  <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>  <span class="comment">!< calculated maximum nsummands</span></div><div class="line"><a name="l00305"></a><span class="lineno"> 305</span> <span class="comment"> !< over all processes</span></div><div class="line"><a name="l00306"></a><span class="lineno"> 306</span> </div><div class="line"><a name="l00307"></a><span class="lineno"> 307</span>  <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>  <span class="comment">!< was total number of summands;</span></div><div class="line"><a name="l00309"></a><span class="lineno"> 309</span> <span class="comment"> !< now is ignored; use </span></div><div class="line"><a name="l00310"></a><span class="lineno"> 310</span> <span class="comment"> !< gbl_max_nsummands instead</span></div><div class="line"><a name="l00311"></a><span class="lineno"> 311</span> </div><div class="line"><a name="l00312"></a><span class="lineno"> 312</span>  <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>  <span class="comment">!< flag enabling/disabling testing that gmax and max_levels are </span></div><div class="line"><a name="l00314"></a><span class="lineno"> 314</span> <span class="comment"> !< accurate/sufficient. Default is enabled.</span></div><div class="line"><a name="l00315"></a><span class="lineno"> 315</span> </div><div class="line"><a name="l00316"></a><span class="lineno"> 316</span>  <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>  <span class="comment">!< increment running totals for</span></div><div class="line"><a name="l00318"></a><span class="lineno"> 318</span> <span class="comment"> !< (1) one-reduction repro_sum</span></div><div class="line"><a name="l00319"></a><span class="lineno"> 319</span> <span class="comment"> !< (2) two-reduction repro_sum</span></div><div class="line"><a name="l00320"></a><span class="lineno"> 320</span> <span class="comment"> !< (3) both types in one call</span></div><div class="line"><a name="l00321"></a><span class="lineno"> 321</span> <span class="comment"> !< (4) nonrepro_sum</span></div><div class="line"><a name="l00322"></a><span class="lineno"> 322</span> <span class="comment"> !< (5) global max nsummands reduction</span></div><div class="line"><a name="l00323"></a><span class="lineno"> 323</span> </div><div class="line"><a name="l00324"></a><span class="lineno"> 324</span>  <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>  <span class="comment">!< relative and absolute</span></div><div class="line"><a name="l00326"></a><span class="lineno"> 326</span> <span class="comment"> !< differences between fixed </span></div><div class="line"><a name="l00327"></a><span class="lineno"> 327</span> <span class="comment"> !< and floating point sums</span></div><div class="line"><a name="l00328"></a><span class="lineno"> 328</span> </div><div class="line"><a name="l00329"></a><span class="lineno"> 329</span>  <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>  <span class="comment">!< MPI communicator</span></div><div class="line"><a name="l00331"></a><span class="lineno"> 331</span> </div><div class="line"><a name="l00332"></a><span class="lineno"> 332</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l00333"></a><span class="lineno"> 333</span> </div><div class="line"><a name="l00334"></a><span class="lineno"> 334</span>  <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>  <span class="comment">! use oasis_reprosum_ddpdd or not</span></div><div class="line"><a name="l00336"></a><span class="lineno"> 336</span>  <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>  <span class="comment">! determine gmax/gmin before </span></div><div class="line"><a name="l00338"></a><span class="lineno"> 338</span>  <span class="comment">! computing sum</span></div><div class="line"><a name="l00339"></a><span class="lineno"> 339</span>  <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>  <span class="comment">! verify gmax and max_levels </span></div><div class="line"><a name="l00341"></a><span class="lineno"> 341</span>  <span class="comment">! are accurate/sufficient</span></div><div class="line"><a name="l00342"></a><span class="lineno"> 342</span>  <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>  <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>  <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>  <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>  <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>  <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>  <span class="comment">! or threads (used in both ways)</span></div><div class="line"><a name="l00349"></a><span class="lineno"> 349</span> </div><div class="line"><a name="l00350"></a><span class="lineno"> 350</span>  <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>  <span class="comment">! range of summand indices for each</span></div><div class="line"><a name="l00352"></a><span class="lineno"> 352</span>  <span class="comment">! OpenMP thread</span></div><div class="line"><a name="l00353"></a><span class="lineno"> 353</span>  <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>  <span class="comment">! per thread local exponent minima</span></div><div class="line"><a name="l00355"></a><span class="lineno"> 355</span>  <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>  <span class="comment">! per thread local exponent maxima</span></div><div class="line"><a name="l00357"></a><span class="lineno"> 357</span>  <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>  <span class="comment">! summand exponent and working min/max</span></div><div class="line"><a name="l00359"></a><span class="lineno"> 359</span>  <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>  <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>  <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>  <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> </div><div class="line"><a name="l00364"></a><span class="lineno"> 364</span>  <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>  <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>  <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>  <span class="comment">! value < 1 (so that sum does</span></div><div class="line"><a name="l00368"></a><span class="lineno"> 368</span>  <span class="comment">! not overflow)</span></div><div class="line"><a name="l00369"></a><span class="lineno"> 369</span>  <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>  <span class="comment">! integer expansion to use</span></div><div class="line"><a name="l00371"></a><span class="lineno"> 371</span>  <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>  <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>  <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>  <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>  <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>  <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> </div><div class="line"><a name="l00378"></a><span class="lineno"> 378</span>  <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>  <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>  <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>  <span class="comment">! fast, nonreproducible, </span></div><div class="line"><a name="l00382"></a><span class="lineno"> 382</span>  <span class="comment">! floating point alg.</span></div><div class="line"><a name="l00383"></a><span class="lineno"> 383</span>  <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>  <span class="comment">! fixed and floating point </span></div><div class="line"><a name="l00385"></a><span class="lineno"> 385</span>  <span class="comment">! sums</span></div><div class="line"><a name="l00386"></a><span class="lineno"> 386</span> <span class="preprocessor">#ifdef _OPENMP</span></div><div class="line"><a name="l00387"></a><span class="lineno"> 387</span> <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>  <span class="keywordtype">external</span> omp_get_max_threads</div><div class="line"><a name="l00389"></a><span class="lineno"> 389</span> <span class="preprocessor">#endif </span></div><div class="line"><a name="l00390"></a><span class="lineno"> 390</span> <span class="preprocessor"></span> <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_calc)'</span></div><div class="line"><a name="l00391"></a><span class="lineno"> 391</span> </div><div class="line"><a name="l00392"></a><span class="lineno"> 392</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00393"></a><span class="lineno"> 393</span> </div><div class="line"><a name="l00394"></a><span class="lineno"> 394</span> <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>  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>  <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>  use_ddpdd_sum = ddpdd_sum</div><div class="line"><a name="l00398"></a><span class="lineno"> 398</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00399"></a><span class="lineno"> 399</span> </div><div class="line"><a name="l00400"></a><span class="lineno"> 400</span> <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> <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> <span class="comment">! If not, always use ddpdd.</span></div><div class="line"><a name="l00403"></a><span class="lineno"> 403</span>  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> </div><div class="line"><a name="l00405"></a><span class="lineno"> 405</span> <span class="comment">! initialize local statistics variables</span></div><div class="line"><a name="l00406"></a><span class="lineno"> 406</span>  gbl_max_red = 0</div><div class="line"><a name="l00407"></a><span class="lineno"> 407</span>  repro_sum_fast = 0</div><div class="line"><a name="l00408"></a><span class="lineno"> 408</span>  repro_sum_slow = 0</div><div class="line"><a name="l00409"></a><span class="lineno"> 409</span>  repro_sum_both = 0</div><div class="line"><a name="l00410"></a><span class="lineno"> 410</span>  nonrepro_sum = 0</div><div class="line"><a name="l00411"></a><span class="lineno"> 411</span> </div><div class="line"><a name="l00412"></a><span class="lineno"> 412</span> <span class="comment">! set MPI communicator</span></div><div class="line"><a name="l00413"></a><span class="lineno"> 413</span>  <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>  mpi_comm = commid</div><div class="line"><a name="l00415"></a><span class="lineno"> 415</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00416"></a><span class="lineno"> 416</span>  mpi_comm = mpi_comm_world</div><div class="line"><a name="l00417"></a><span class="lineno"> 417</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00418"></a><span class="lineno"> 418</span> </div><div class="line"><a name="l00419"></a><span class="lineno"> 419</span>  <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>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">'oasis_reprosum_prebarrier'</span>)</div><div class="line"><a name="l00421"></a><span class="lineno"> 421</span>  <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>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a> (<span class="stringliteral">'oasis_reprosum_prebarrier'</span>)</div><div class="line"><a name="l00423"></a><span class="lineno"> 423</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00424"></a><span class="lineno"> 424</span> </div><div class="line"><a name="l00425"></a><span class="lineno"> 425</span>  <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> </div><div class="line"><a name="l00427"></a><span class="lineno"> 427</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">'oasis_reprosum_ddpdd'</span>)</div><div class="line"><a name="l00428"></a><span class="lineno"> 428</span> </div><div class="line"><a name="l00429"></a><span class="lineno"> 429</span>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">oasis_reprosum_ddpdd</a>(arr, arr_gsum, nsummands, dsummands, &</div><div class="line"><a name="l00430"></a><span class="lineno"> 430</span>  nflds, mpi_comm)</div><div class="line"><a name="l00431"></a><span class="lineno"> 431</span>  repro_sum_fast = 1</div><div class="line"><a name="l00432"></a><span class="lineno"> 432</span> </div><div class="line"><a name="l00433"></a><span class="lineno"> 433</span>  <span class="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">'oasis_reprosum_ddpdd'</span>)</div><div class="line"><a name="l00434"></a><span class="lineno"> 434</span> </div><div class="line"><a name="l00435"></a><span class="lineno"> 435</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00436"></a><span class="lineno"> 436</span> </div><div class="line"><a name="l00437"></a><span class="lineno"> 437</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">'oasis_reprosum_int'</span>)</div><div class="line"><a name="l00438"></a><span class="lineno"> 438</span> </div><div class="line"><a name="l00439"></a><span class="lineno"> 439</span> <span class="comment">! get number of MPI tasks</span></div><div class="line"><a name="l00440"></a><span class="lineno"> 440</span>  <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> </div><div class="line"><a name="l00442"></a><span class="lineno"> 442</span> <span class="comment">! get number of OpenMP threads</span></div><div class="line"><a name="l00443"></a><span class="lineno"> 443</span> <span class="preprocessor">#ifdef _OPENMP</span></div><div class="line"><a name="l00444"></a><span class="lineno"> 444</span> <span class="preprocessor"></span> omp_nthreads = omp_get_max_threads()</div><div class="line"><a name="l00445"></a><span class="lineno"> 445</span> <span class="preprocessor">#else</span></div><div class="line"><a name="l00446"></a><span class="lineno"> 446</span> <span class="preprocessor"></span> omp_nthreads = 1</div><div class="line"><a name="l00447"></a><span class="lineno"> 447</span> <span class="preprocessor">#endif</span></div><div class="line"><a name="l00448"></a><span class="lineno"> 448</span> <span class="preprocessor"></span></div><div class="line"><a name="l00449"></a><span class="lineno"> 449</span> <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>  recompute = .true.</div><div class="line"><a name="l00451"></a><span class="lineno"> 451</span>  validate = .false.</div><div class="line"><a name="l00452"></a><span class="lineno"> 452</span>  <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>  recompute = .false.</div><div class="line"><a name="l00454"></a><span class="lineno"> 454</span> </div><div class="line"><a name="l00455"></a><span class="lineno"> 455</span> <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> <span class="comment">! performance for loopb in oasis_reprosum_int</span></div><div class="line"><a name="l00457"></a><span class="lineno"> 457</span>  max_level = (64/nflds) + 1</div><div class="line"><a name="l00458"></a><span class="lineno"> 458</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00459"></a><span class="lineno"> 459</span>  <span class="keywordflow">if</span> ((arr_gbl_max(ifld) .ge. 0.0_r8) .and. &</div><div class="line"><a name="l00460"></a><span class="lineno"> 460</span>  (arr_max_levels(ifld) > 0)) <span class="keywordflow">then</span></div><div class="line"><a name="l00461"></a><span class="lineno"> 461</span> </div><div class="line"><a name="l00462"></a><span class="lineno"> 462</span>  arr_gmax_exp(ifld) = exponent(arr_gbl_max(ifld))</div><div class="line"><a name="l00463"></a><span class="lineno"> 463</span>  <span class="keywordflow">if</span> (max_level < arr_max_levels(ifld)) &</div><div class="line"><a name="l00464"></a><span class="lineno"> 464</span>  max_level = arr_max_levels(ifld)</div><div class="line"><a name="l00465"></a><span class="lineno"> 465</span> </div><div class="line"><a name="l00466"></a><span class="lineno"> 466</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00467"></a><span class="lineno"> 467</span>  recompute = .true.</div><div class="line"><a name="l00468"></a><span class="lineno"> 468</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00469"></a><span class="lineno"> 469</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00470"></a><span class="lineno"> 470</span> </div><div class="line"><a name="l00471"></a><span class="lineno"> 471</span>  <span class="keywordflow">if</span> (.not. recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00472"></a><span class="lineno"> 472</span> </div><div class="line"><a name="l00473"></a><span class="lineno"> 473</span> <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> <span class="comment">! algorithm</span></div><div class="line"><a name="l00475"></a><span class="lineno"> 475</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">"repro_sum_allr_max"</span>)</div><div class="line"><a name="l00476"></a><span class="lineno"> 476</span>  <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>  <span class="keywordflow">if</span> (gbl_max_nsummands < 1) <span class="keywordflow">then</span></div><div class="line"><a name="l00478"></a><span class="lineno"> 478</span>  <span class="keyword">call </span>mpi_allreduce (nsummands, max_nsummands, 1, &</div><div class="line"><a name="l00479"></a><span class="lineno"> 479</span>  mpi_integer, mpi_max, mpi_comm, ierr)</div><div class="line"><a name="l00480"></a><span class="lineno"> 480</span>  gbl_max_red = 1</div><div class="line"><a name="l00481"></a><span class="lineno"> 481</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00482"></a><span class="lineno"> 482</span>  max_nsummands = gbl_max_nsummands</div><div class="line"><a name="l00483"></a><span class="lineno"> 483</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00484"></a><span class="lineno"> 484</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00485"></a><span class="lineno"> 485</span>  <span class="keyword">call </span>mpi_allreduce (nsummands, max_nsummands, 1, &</div><div class="line"><a name="l00486"></a><span class="lineno"> 486</span>  mpi_integer, mpi_max, mpi_comm, ierr)</div><div class="line"><a name="l00487"></a><span class="lineno"> 487</span>  gbl_max_red = 1</div><div class="line"><a name="l00488"></a><span class="lineno"> 488</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00489"></a><span class="lineno"> 489</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">"repro_sum_allr_max"</span>)</div><div class="line"><a name="l00490"></a><span class="lineno"> 490</span> </div><div class="line"><a name="l00491"></a><span class="lineno"> 491</span> <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> <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> </div><div class="line"><a name="l00494"></a><span class="lineno"> 494</span> <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>  <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>  gbl_max_nsummands_out = max_nsummands</div><div class="line"><a name="l00497"></a><span class="lineno"> 497</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00498"></a><span class="lineno"> 498</span> </div><div class="line"><a name="l00499"></a><span class="lineno"> 499</span> <span class="comment">! summing within each thread first</span></div><div class="line"><a name="l00500"></a><span class="lineno"> 500</span>  max_nsummands = (max_nsummands/omp_nthreads) + 1</div><div class="line"><a name="l00501"></a><span class="lineno"> 501</span> <span class="comment">! then over threads and tasks</span></div><div class="line"><a name="l00502"></a><span class="lineno"> 502</span>  max_nsummands = max(max_nsummands, tasks*omp_nthreads)</div><div class="line"><a name="l00503"></a><span class="lineno"> 503</span> </div><div class="line"><a name="l00504"></a><span class="lineno"> 504</span>  xmax_nsummands = dble(max_nsummands)</div><div class="line"><a name="l00505"></a><span class="lineno"> 505</span>  arr_max_shift = digits(0_i8) - (exponent(xmax_nsummands) + 1)</div><div class="line"><a name="l00506"></a><span class="lineno"> 506</span>  <span class="keywordflow">if</span> (arr_max_shift < 2) <span class="keywordflow">then</span></div><div class="line"><a name="l00507"></a><span class="lineno"> 507</span>  <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">' number of summands too large for fixed precision algorithm'</span></div><div class="line"><a name="l00508"></a><span class="lineno"> 508</span>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00510"></a><span class="lineno"> 510</span> </div><div class="line"><a name="l00511"></a><span class="lineno"> 511</span> <span class="comment">! calculate sum</span></div><div class="line"><a name="l00512"></a><span class="lineno"> 512</span>  <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>  validate = repro_sum_validate</div><div class="line"><a name="l00514"></a><span class="lineno"> 514</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00515"></a><span class="lineno"> 515</span>  validate = .true.</div><div class="line"><a name="l00516"></a><span class="lineno"> 516</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00517"></a><span class="lineno"> 517</span>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a>(arr, arr_gsum, nsummands, dsummands, &</div><div class="line"><a name="l00518"></a><span class="lineno"> 518</span>  nflds, arr_max_shift, arr_gmax_exp, &</div><div class="line"><a name="l00519"></a><span class="lineno"> 519</span>  arr_max_levels, max_level, validate, &</div><div class="line"><a name="l00520"></a><span class="lineno"> 520</span>  recompute, omp_nthreads, mpi_comm)</div><div class="line"><a name="l00521"></a><span class="lineno"> 521</span> </div><div class="line"><a name="l00522"></a><span class="lineno"> 522</span> <span class="comment">! record statistics, etc.</span></div><div class="line"><a name="l00523"></a><span class="lineno"> 523</span>  repro_sum_fast = 1</div><div class="line"><a name="l00524"></a><span class="lineno"> 524</span>  <span class="keywordflow">if</span> (recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00525"></a><span class="lineno"> 525</span>  repro_sum_both = 1</div><div class="line"><a name="l00526"></a><span class="lineno"> 526</span>  <span class="keywordflow">else</span> </div><div class="line"><a name="l00527"></a><span class="lineno"> 527</span> <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>  <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>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00530"></a><span class="lineno"> 530</span>  arr_max_levels_out(ifld) = arr_max_levels(ifld)</div><div class="line"><a name="l00531"></a><span class="lineno"> 531</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00532"></a><span class="lineno"> 532</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00533"></a><span class="lineno"> 533</span>  <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>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00535"></a><span class="lineno"> 535</span>  arr_gbl_max_out(ifld) = arr_gbl_max(ifld)</div><div class="line"><a name="l00536"></a><span class="lineno"> 536</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00537"></a><span class="lineno"> 537</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00538"></a><span class="lineno"> 538</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00539"></a><span class="lineno"> 539</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00540"></a><span class="lineno"> 540</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00541"></a><span class="lineno"> 541</span> </div><div class="line"><a name="l00542"></a><span class="lineno"> 542</span> <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> <span class="comment">! use to compute required number of levels</span></div><div class="line"><a name="l00544"></a><span class="lineno"> 544</span>  <span class="keywordflow">if</span> (recompute) <span class="keywordflow">then</span></div><div class="line"><a name="l00545"></a><span class="lineno"> 545</span> </div><div class="line"><a name="l00546"></a><span class="lineno"> 546</span> <span class="comment">! record statistic</span></div><div class="line"><a name="l00547"></a><span class="lineno"> 547</span>  repro_sum_slow = 1</div><div class="line"><a name="l00548"></a><span class="lineno"> 548</span> </div><div class="line"><a name="l00549"></a><span class="lineno"> 549</span> <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> <span class="comment">! maximum number of local summands</span></div><div class="line"><a name="l00551"></a><span class="lineno"> 551</span> </div><div class="line"><a name="l00552"></a><span class="lineno"> 552</span> <span class="comment">! allocate thread-specific work space</span></div><div class="line"><a name="l00553"></a><span class="lineno"> 553</span>  <span class="keyword">allocate</span>(arr_tlmax_exp(nflds,omp_nthreads))</div><div class="line"><a name="l00554"></a><span class="lineno"> 554</span>  <span class="keyword">allocate</span>(arr_tlmin_exp(nflds,omp_nthreads))</div><div class="line"><a name="l00555"></a><span class="lineno"> 555</span>  <span class="keyword">allocate</span>(isum_beg(omp_nthreads))</div><div class="line"><a name="l00556"></a><span class="lineno"> 556</span>  <span class="keyword">allocate</span>(isum_end(omp_nthreads))</div><div class="line"><a name="l00557"></a><span class="lineno"> 557</span> </div><div class="line"><a name="l00558"></a><span class="lineno"> 558</span> <span class="comment">! split summand index range over OpenMP threads</span></div><div class="line"><a name="l00559"></a><span class="lineno"> 559</span>  <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> </div><div class="line"><a name="l00561"></a><span class="lineno"> 561</span> <span class="comment">!$omp parallel do &</span></div><div class="line"><a name="l00562"></a><span class="lineno"> 562</span> <span class="comment">!$omp default(shared) &</span></div><div class="line"><a name="l00563"></a><span class="lineno"> 563</span> <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>  <span class="keywordflow">do</span> ithread=1,omp_nthreads</div><div class="line"><a name="l00565"></a><span class="lineno"> 565</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">'repro_sum_loopa'</span>)</div><div class="line"><a name="l00566"></a><span class="lineno"> 566</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00567"></a><span class="lineno"> 567</span>  arr_exp_tlmin = maxexponent(1._r8)</div><div class="line"><a name="l00568"></a><span class="lineno"> 568</span>  arr_exp_tlmax = minexponent(1._r8)</div><div class="line"><a name="l00569"></a><span class="lineno"> 569</span>  <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>  <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>  arr_exp = exponent(arr(isum,ifld))</div><div class="line"><a name="l00572"></a><span class="lineno"> 572</span>  arr_exp_tlmin = min(arr_exp,arr_exp_tlmin)</div><div class="line"><a name="l00573"></a><span class="lineno"> 573</span>  arr_exp_tlmax = max(arr_exp,arr_exp_tlmax)</div><div class="line"><a name="l00574"></a><span class="lineno"> 574</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00575"></a><span class="lineno"> 575</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00576"></a><span class="lineno"> 576</span>  arr_tlmin_exp(ifld,ithread) = arr_exp_tlmin</div><div class="line"><a name="l00577"></a><span class="lineno"> 577</span>  arr_tlmax_exp(ifld,ithread) = arr_exp_tlmax</div><div class="line"><a name="l00578"></a><span class="lineno"> 578</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00579"></a><span class="lineno"> 579</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">'repro_sum_loopa'</span>)</div><div class="line"><a name="l00580"></a><span class="lineno"> 580</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00581"></a><span class="lineno"> 581</span> </div><div class="line"><a name="l00582"></a><span class="lineno"> 582</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00583"></a><span class="lineno"> 583</span>  arr_lmax_exp(ifld) = maxval(arr_tlmax_exp(ifld,:))</div><div class="line"><a name="l00584"></a><span class="lineno"> 584</span>  arr_lmin_exp(ifld) = minval(arr_tlmin_exp(ifld,:))</div><div class="line"><a name="l00585"></a><span class="lineno"> 585</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00586"></a><span class="lineno"> 586</span>  <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> </div><div class="line"><a name="l00588"></a><span class="lineno"> 588</span>  arr_lextremes(0,:) = -nsummands</div><div class="line"><a name="l00589"></a><span class="lineno"> 589</span>  arr_lextremes(1:nflds,1) = -arr_lmax_exp(:)</div><div class="line"><a name="l00590"></a><span class="lineno"> 590</span>  arr_lextremes(1:nflds,2) = arr_lmin_exp(:)</div><div class="line"><a name="l00591"></a><span class="lineno"> 591</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">"repro_sum_allr_minmax"</span>)</div><div class="line"><a name="l00592"></a><span class="lineno"> 592</span>  <span class="keyword">call </span>mpi_allreduce (arr_lextremes, arr_gextremes, 2*(nflds+1), &</div><div class="line"><a name="l00593"></a><span class="lineno"> 593</span>  mpi_integer, mpi_min, mpi_comm, ierr)</div><div class="line"><a name="l00594"></a><span class="lineno"> 594</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">"repro_sum_allr_minmax"</span>)</div><div class="line"><a name="l00595"></a><span class="lineno"> 595</span>  max_nsummands = -arr_gextremes(0,1)</div><div class="line"><a name="l00596"></a><span class="lineno"> 596</span>  arr_gmax_exp(:) = -arr_gextremes(1:nflds,1)</div><div class="line"><a name="l00597"></a><span class="lineno"> 597</span>  arr_gmin_exp(:) = arr_gextremes(1:nflds,2)</div><div class="line"><a name="l00598"></a><span class="lineno"> 598</span> </div><div class="line"><a name="l00599"></a><span class="lineno"> 599</span> <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> <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> <span class="comment">! arr_gmin_exp = arr_gmax_exp = MINEXPONENT</span></div><div class="line"><a name="l00602"></a><span class="lineno"> 602</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00603"></a><span class="lineno"> 603</span>  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> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00605"></a><span class="lineno"> 605</span> </div><div class="line"><a name="l00606"></a><span class="lineno"> 606</span> <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>  <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>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00609"></a><span class="lineno"> 609</span>  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> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00611"></a><span class="lineno"> 611</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00612"></a><span class="lineno"> 612</span> </div><div class="line"><a name="l00613"></a><span class="lineno"> 613</span> <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>  <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>  gbl_max_nsummands_out = max_nsummands</div><div class="line"><a name="l00616"></a><span class="lineno"> 616</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00617"></a><span class="lineno"> 617</span> </div><div class="line"><a name="l00618"></a><span class="lineno"> 618</span> <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> <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> <span class="comment">! exceed maximum number of digits in i8.</span></div><div class="line"><a name="l00621"></a><span class="lineno"> 621</span> </div><div class="line"><a name="l00622"></a><span class="lineno"> 622</span> <span class="comment">! summing within each thread first</span></div><div class="line"><a name="l00623"></a><span class="lineno"> 623</span>  max_nsummands = (max_nsummands/omp_nthreads) + 1</div><div class="line"><a name="l00624"></a><span class="lineno"> 624</span> <span class="comment">! then over threads and tasks</span></div><div class="line"><a name="l00625"></a><span class="lineno"> 625</span>  max_nsummands = max(max_nsummands, tasks*omp_nthreads)</div><div class="line"><a name="l00626"></a><span class="lineno"> 626</span> </div><div class="line"><a name="l00627"></a><span class="lineno"> 627</span>  xmax_nsummands = dble(max_nsummands)</div><div class="line"><a name="l00628"></a><span class="lineno"> 628</span>  arr_max_shift = digits(0_i8) - (exponent(xmax_nsummands) + 1)</div><div class="line"><a name="l00629"></a><span class="lineno"> 629</span>  <span class="keywordflow">if</span> (arr_max_shift < 2) <span class="keywordflow">then</span></div><div class="line"><a name="l00630"></a><span class="lineno"> 630</span>  <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">' number of summands too large for fixed precision algorithm'</span></div><div class="line"><a name="l00631"></a><span class="lineno"> 631</span>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00633"></a><span class="lineno"> 633</span> </div><div class="line"><a name="l00634"></a><span class="lineno"> 634</span> <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> <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> <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> <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> <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> <span class="comment">! performance for loopb in oasis_reprosum_int)</span></div><div class="line"><a name="l00640"></a><span class="lineno"> 640</span>  max_level = (64/nflds) + 1</div><div class="line"><a name="l00641"></a><span class="lineno"> 641</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00642"></a><span class="lineno"> 642</span>  max_levels(ifld) = 2 + &</div><div class="line"><a name="l00643"></a><span class="lineno"> 643</span>  ((digits(0_i8) + (arr_gmax_exp(ifld)-arr_gmin_exp(ifld))) &</div><div class="line"><a name="l00644"></a><span class="lineno"> 644</span>  / arr_max_shift)</div><div class="line"><a name="l00645"></a><span class="lineno"> 645</span>  <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> <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> <span class="comment">! previously</span></div><div class="line"><a name="l00648"></a><span class="lineno"> 648</span>  <span class="keywordflow">if</span> ( arr_max_levels(ifld) > 0 ) <span class="keywordflow">then</span></div><div class="line"><a name="l00649"></a><span class="lineno"> 649</span>  max_levels(ifld) = &</div><div class="line"><a name="l00650"></a><span class="lineno"> 650</span>  min(arr_max_levels(ifld),max_levels(ifld))</div><div class="line"><a name="l00651"></a><span class="lineno"> 651</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00652"></a><span class="lineno"> 652</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00653"></a><span class="lineno"> 653</span>  <span class="keywordflow">if</span> (max_level < max_levels(ifld)) &</div><div class="line"><a name="l00654"></a><span class="lineno"> 654</span>  max_level = max_levels(ifld)</div><div class="line"><a name="l00655"></a><span class="lineno"> 655</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00656"></a><span class="lineno"> 656</span> </div><div class="line"><a name="l00657"></a><span class="lineno"> 657</span> <span class="comment">! if requested, return calculated levels</span></div><div class="line"><a name="l00658"></a><span class="lineno"> 658</span>  <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>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00660"></a><span class="lineno"> 660</span>  arr_max_levels_out(ifld) = max_levels(ifld)</div><div class="line"><a name="l00661"></a><span class="lineno"> 661</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00662"></a><span class="lineno"> 662</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00663"></a><span class="lineno"> 663</span> </div><div class="line"><a name="l00664"></a><span class="lineno"> 664</span> <span class="comment">! calculate sum</span></div><div class="line"><a name="l00665"></a><span class="lineno"> 665</span>  validate = .false.</div><div class="line"><a name="l00666"></a><span class="lineno"> 666</span>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a>(arr, arr_gsum, nsummands, dsummands, nflds, &</div><div class="line"><a name="l00667"></a><span class="lineno"> 667</span>  arr_max_shift, arr_gmax_exp, max_levels, &</div><div class="line"><a name="l00668"></a><span class="lineno"> 668</span>  max_level, validate, recompute, &</div><div class="line"><a name="l00669"></a><span class="lineno"> 669</span>  omp_nthreads, mpi_comm)</div><div class="line"><a name="l00670"></a><span class="lineno"> 670</span> </div><div class="line"><a name="l00671"></a><span class="lineno"> 671</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00672"></a><span class="lineno"> 672</span> </div><div class="line"><a name="l00673"></a><span class="lineno"> 673</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">'oasis_reprosum_int'</span>)</div><div class="line"><a name="l00674"></a><span class="lineno"> 674</span> </div><div class="line"><a name="l00675"></a><span class="lineno"> 675</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00676"></a><span class="lineno"> 676</span> </div><div class="line"><a name="l00677"></a><span class="lineno"> 677</span> <span class="comment">! compare fixed and floating point results</span></div><div class="line"><a name="l00678"></a><span class="lineno"> 678</span>  <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>  <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> >= 0.0_r8) <span class="keywordflow">then</span></div><div class="line"><a name="l00680"></a><span class="lineno"> 680</span> </div><div class="line"><a name="l00681"></a><span class="lineno"> 681</span>  <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>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a13a16c07778911b45f3f42a542348019">oasis_timer_start</a>(<span class="stringliteral">'oasis_nonreprosum_prebarrier'</span>)</div><div class="line"><a name="l00683"></a><span class="lineno"> 683</span>  <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>  <span class="keyword">call </span><a class="code" href="namespacemod__oasis__timer.html#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a> (<span class="stringliteral">'oasis_nonreprosum_prebarrier'</span>)</div><div class="line"><a name="l00685"></a><span class="lineno"> 685</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00686"></a><span class="lineno"> 686</span> </div><div class="line"><a name="l00687"></a><span class="lineno"> 687</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">'nonrepro_sum'</span>)</div><div class="line"><a name="l00688"></a><span class="lineno"> 688</span> <span class="comment">! record statistic</span></div><div class="line"><a name="l00689"></a><span class="lineno"> 689</span>  nonrepro_sum = 1</div><div class="line"><a name="l00690"></a><span class="lineno"> 690</span> <span class="comment">! compute nonreproducible sum</span></div><div class="line"><a name="l00691"></a><span class="lineno"> 691</span>  arr_lsum(:) = 0._r8</div><div class="line"><a name="l00692"></a><span class="lineno"> 692</span> <span class="comment">!$omp parallel do &</span></div><div class="line"><a name="l00693"></a><span class="lineno"> 693</span> <span class="comment">!$omp default(shared) &</span></div><div class="line"><a name="l00694"></a><span class="lineno"> 694</span> <span class="comment">!$omp private(ifld, isum)</span></div><div class="line"><a name="l00695"></a><span class="lineno"> 695</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00696"></a><span class="lineno"> 696</span>  <span class="keywordflow">do</span> isum=1,nsummands</div><div class="line"><a name="l00697"></a><span class="lineno"> 697</span>  arr_lsum(ifld) = arr(isum,ifld) + arr_lsum(ifld)</div><div class="line"><a name="l00698"></a><span class="lineno"> 698</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00699"></a><span class="lineno"> 699</span> <span class="keywordflow"> end do</span></div><div class="line"><a name="l00700"></a><span class="lineno"> 700</span> </div><div class="line"><a name="l00701"></a><span class="lineno"> 701</span>  <span class="keyword">call </span>mpi_allreduce (arr_lsum, arr_gsum_fast, nflds, &</div><div class="line"><a name="l00702"></a><span class="lineno"> 702</span>  mpi_real8, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l00703"></a><span class="lineno"> 703</span> </div><div class="line"><a name="l00704"></a><span class="lineno"> 704</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">'nonrepro_sum'</span>)</div><div class="line"><a name="l00705"></a><span class="lineno"> 705</span> </div><div class="line"><a name="l00706"></a><span class="lineno"> 706</span> <span class="comment">! determine differences</span></div><div class="line"><a name="l00707"></a><span class="lineno"> 707</span> <span class="comment">!$omp parallel do &</span></div><div class="line"><a name="l00708"></a><span class="lineno"> 708</span> <span class="comment">!$omp default(shared) &</span></div><div class="line"><a name="l00709"></a><span class="lineno"> 709</span> <span class="comment">!$omp private(ifld, abs_diff)</span></div><div class="line"><a name="l00710"></a><span class="lineno"> 710</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00711"></a><span class="lineno"> 711</span>  abs_diff = abs(arr_gsum_fast(ifld)-arr_gsum(ifld))</div><div class="line"><a name="l00712"></a><span class="lineno"> 712</span>  <span class="keywordflow">if</span> (abs(arr_gsum(ifld)) > abs_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l00713"></a><span class="lineno"> 713</span>  rel_diff(1,ifld) = abs_diff/abs(arr_gsum(ifld))</div><div class="line"><a name="l00714"></a><span class="lineno"> 714</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00715"></a><span class="lineno"> 715</span>  rel_diff(1,ifld) = abs_diff</div><div class="line"><a name="l00716"></a><span class="lineno"> 716</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00717"></a><span class="lineno"> 717</span>  rel_diff(2,ifld) = abs_diff</div><div class="line"><a name="l00718"></a><span class="lineno"> 718</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00719"></a><span class="lineno"> 719</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00720"></a><span class="lineno"> 720</span>  rel_diff(:,:) = 0.0_r8</div><div class="line"><a name="l00721"></a><span class="lineno"> 721</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00722"></a><span class="lineno"> 722</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00723"></a><span class="lineno"> 723</span> </div><div class="line"><a name="l00724"></a><span class="lineno"> 724</span> <span class="comment">! return statistics</span></div><div class="line"><a name="l00725"></a><span class="lineno"> 725</span>  <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>  repro_sum_stats(1) = repro_sum_stats(1) + repro_sum_fast</div><div class="line"><a name="l00727"></a><span class="lineno"> 727</span>  repro_sum_stats(2) = repro_sum_stats(2) + repro_sum_slow</div><div class="line"><a name="l00728"></a><span class="lineno"> 728</span>  repro_sum_stats(3) = repro_sum_stats(3) + repro_sum_both</div><div class="line"><a name="l00729"></a><span class="lineno"> 729</span>  repro_sum_stats(4) = repro_sum_stats(4) + nonrepro_sum</div><div class="line"><a name="l00730"></a><span class="lineno"> 730</span>  repro_sum_stats(5) = repro_sum_stats(5) + gbl_max_red</div><div class="line"><a name="l00731"></a><span class="lineno"> 731</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00732"></a><span class="lineno"> 732</span> </div><div class="line"><a name="l00733"></a><span class="lineno"> 733</span> </div><div class="line"><a name="l00734"></a><span class="lineno"> 734</span> <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> </div><div class="line"><a name="l00736"></a><span class="lineno"> 736</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l00737"></a><span class="lineno"> 737</span> <span class="comment">! </span></div><div class="line"><a name="l00738"></a><span class="lineno"> 738</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l00739"></a><span class="lineno"> 739</span> <span class="comment">!> Compute the global sum of each field in "arr" using the indicated </span></div><div class="line"><a name="l00740"></a><span class="lineno"> 740</span> <span class="comment">!> communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l00741"></a><span class="lineno"> 741</span> <span class="comment">!> 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> <span class="comment">!> is controlled by the number of "levels" of integer expansion, the</span></div><div class="line"><a name="l00743"></a><span class="lineno"> 743</span> <span class="comment">!> maximum value of which is specified by max_level. </span></div><div class="line"><a name="l00744"></a><span class="lineno"> 744</span> <span class="comment">! </span></div><div class="line"><a name="l00745"></a><span class="lineno"> 745</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00746"></a><span class="lineno"> 746</span> </div><div class="line"><a name="l00747"></a><span class="lineno"> 747</span> <span class="keyword"> subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#a7ea3c5bad9db10c80eb528453602aed3">oasis_reprosum_int</a> (arr, arr_gsum, nsummands, dsummands, nflds, &</div><div class="line"><a name="l00748"></a><span class="lineno"> 748</span>  arr_max_shift, arr_gmax_exp, max_levels, &</div><div class="line"><a name="l00749"></a><span class="lineno"> 749</span>  max_level, validate, recompute, &</div><div class="line"><a name="l00750"></a><span class="lineno"> 750</span>  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> </div><div class="line"><a name="l00752"></a><span class="lineno"> 752</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l00753"></a><span class="lineno"> 753</span> </div><div class="line"><a name="l00754"></a><span class="lineno"> 754</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l00755"></a><span class="lineno"> 755</span> </div><div class="line"><a name="l00756"></a><span class="lineno"> 756</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nsummands<span class="comment"> !< number of local summands</span></div><div class="line"><a name="l00757"></a><span class="lineno"> 757</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: dsummands<span class="comment"> !< declared first dimension</span></div><div class="line"><a name="l00758"></a><span class="lineno"> 758</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nflds<span class="comment"> !< number of fields</span></div><div class="line"><a name="l00759"></a><span class="lineno"> 759</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: arr_max_shift<span class="comment"> !< maximum safe exponent for </span></div><div class="line"><a name="l00760"></a><span class="lineno"> 760</span> <span class="comment"> !< value < 1 (so that sum </span></div><div class="line"><a name="l00761"></a><span class="lineno"> 761</span> <span class="comment"> !< does not overflow)</span></div><div class="line"><a name="l00762"></a><span class="lineno"> 762</span>  <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>  <span class="comment">!< exponents of global maxima</span></div><div class="line"><a name="l00764"></a><span class="lineno"> 764</span>  <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>  <span class="comment">!< maximum number of levels </span></div><div class="line"><a name="l00766"></a><span class="lineno"> 766</span> <span class="comment"> !< of integer expansion</span></div><div class="line"><a name="l00767"></a><span class="lineno"> 767</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: max_level<span class="comment"> !< maximum value in </span></div><div class="line"><a name="l00768"></a><span class="lineno"> 768</span> <span class="comment"> !< max_levels</span></div><div class="line"><a name="l00769"></a><span class="lineno"> 769</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: omp_nthreads<span class="comment"> !< number of OpenMP threads</span></div><div class="line"><a name="l00770"></a><span class="lineno"> 770</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: mpi_comm<span class="comment"> !< MPI subcommunicator</span></div><div class="line"><a name="l00771"></a><span class="lineno"> 771</span> </div><div class="line"><a name="l00772"></a><span class="lineno"> 772</span>  <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>  <span class="comment">!< input array</span></div><div class="line"><a name="l00774"></a><span class="lineno"> 774</span> </div><div class="line"><a name="l00775"></a><span class="lineno"> 775</span>  <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>  <span class="comment">!< flag indicating that accuracy of solution generated from</span></div><div class="line"><a name="l00777"></a><span class="lineno"> 777</span> <span class="comment"> !< arr_gmax_exp and max_levels should be tested</span></div><div class="line"><a name="l00778"></a><span class="lineno"> 778</span> </div><div class="line"><a name="l00779"></a><span class="lineno"> 779</span>  <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>  <span class="comment">!< flag indicating that either the upper bounds are inaccurate,</span></div><div class="line"><a name="l00781"></a><span class="lineno"> 781</span> <span class="comment"> !< 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> <span class="comment"> !< enough sums</span></div><div class="line"><a name="l00783"></a><span class="lineno"> 783</span> </div><div class="line"><a name="l00784"></a><span class="lineno"> 784</span>  <span class="keywordtype">real(r8)</span>, <span class="keywordtype">intent(out)</span>:: arr_gsum(nflds)<span class="comment"> !< global means</span></div><div class="line"><a name="l00785"></a><span class="lineno"> 785</span> </div><div class="line"><a name="l00786"></a><span class="lineno"> 786</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l00787"></a><span class="lineno"> 787</span> </div><div class="line"><a name="l00788"></a><span class="lineno"> 788</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">parameter</span> :: max_jlevel = &</div><div class="line"><a name="l00789"></a><span class="lineno"> 789</span>  1 + (digits(0_i8)/digits(0.0_r8))</div><div class="line"><a name="l00790"></a><span class="lineno"> 790</span> </div><div class="line"><a name="l00791"></a><span class="lineno"> 791</span>  <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>  <span class="comment">! integer vector representing local </span></div><div class="line"><a name="l00793"></a><span class="lineno"> 793</span>  <span class="comment">! sum (per thread, per field)</span></div><div class="line"><a name="l00794"></a><span class="lineno"> 794</span>  <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>  <span class="comment">! integer vector representing local </span></div><div class="line"><a name="l00796"></a><span class="lineno"> 796</span>  <span class="comment">! sum</span></div><div class="line"><a name="l00797"></a><span class="lineno"> 797</span>  <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>  <span class="comment">! expansion level</span></div><div class="line"><a name="l00799"></a><span class="lineno"> 799</span>  <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>  <span class="comment">! integer vector representing global</span></div><div class="line"><a name="l00801"></a><span class="lineno"> 801</span>  <span class="comment">! sum</span></div><div class="line"><a name="l00802"></a><span class="lineno"> 802</span>  <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>  <span class="comment">! jlevels of X_8 ('part' of </span></div><div class="line"><a name="l00804"></a><span class="lineno"> 804</span>  <span class="comment">! i8_arr_gsum_level)</span></div><div class="line"><a name="l00805"></a><span class="lineno"> 805</span>  <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>  <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> </div><div class="line"><a name="l00808"></a><span class="lineno"> 808</span>  <span class="keywordtype">integer</span> :: max_error(nflds,omp_nthreads) </div><div class="line"><a name="l00809"></a><span class="lineno"> 809</span>  <span class="comment">! accurate upper bound on data?</span></div><div class="line"><a name="l00810"></a><span class="lineno"> 810</span>  <span class="keywordtype">integer</span> :: not_exact(nflds,omp_nthreads) </div><div class="line"><a name="l00811"></a><span class="lineno"> 811</span>  <span class="comment">! max_levels sufficient to </span></div><div class="line"><a name="l00812"></a><span class="lineno"> 812</span>  <span class="comment">! capture all digits?</span></div><div class="line"><a name="l00813"></a><span class="lineno"> 813</span>  <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>  <span class="comment">! range of summand indices for each</span></div><div class="line"><a name="l00815"></a><span class="lineno"> 815</span>  <span class="comment">! OpenMP thread</span></div><div class="line"><a name="l00816"></a><span class="lineno"> 816</span>  <span class="keywordtype">integer</span> :: ifld, isum, ithread</div><div class="line"><a name="l00817"></a><span class="lineno"> 817</span>  <span class="comment">! loop variables</span></div><div class="line"><a name="l00818"></a><span class="lineno"> 818</span>  <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>  <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>  <span class="comment">! for current expansion level</span></div><div class="line"><a name="l00821"></a><span class="lineno"> 821</span>  <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>  <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>  <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>  <span class="comment">! expansion of current ifld</span></div><div class="line"><a name="l00825"></a><span class="lineno"> 825</span>  <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>  <span class="comment">! include validation metrics </span></div><div class="line"><a name="l00827"></a><span class="lineno"> 827</span>  <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>  <span class="keywordtype">integer</span> :: jlevel <span class="comment">! number of floating point 'pieces' </span></div><div class="line"><a name="l00829"></a><span class="lineno"> 829</span>  <span class="comment">! extracted from a given i8 integer</span></div><div class="line"><a name="l00830"></a><span class="lineno"> 830</span>  <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>  <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>  <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>  <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>  <span class="comment">! in integer expansion of sum</span></div><div class="line"><a name="l00835"></a><span class="lineno"> 835</span>  <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>  <span class="comment">! reconstruction from integer vector</span></div><div class="line"><a name="l00837"></a><span class="lineno"> 837</span>  <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>  <span class="comment">! reconstruction from integer vector</span></div><div class="line"><a name="l00839"></a><span class="lineno"> 839</span> </div><div class="line"><a name="l00840"></a><span class="lineno"> 840</span>  <span class="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>  <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>  <span class="comment">! current level of integer expansion</span></div><div class="line"><a name="l00843"></a><span class="lineno"> 843</span>  <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>  <span class="comment">! i8_arr_gsum_level</span></div><div class="line"><a name="l00845"></a><span class="lineno"> 845</span>  <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>  <span class="comment">! between current i8_arr_gsum_level</span></div><div class="line"><a name="l00847"></a><span class="lineno"> 847</span>  <span class="comment">! and current jlevels of X_8 </span></div><div class="line"><a name="l00848"></a><span class="lineno"> 848</span>  <span class="comment">! (== IX_8). Also used in final</span></div><div class="line"><a name="l00849"></a><span class="lineno"> 849</span>  <span class="comment">! scaling step</span></div><div class="line"><a name="l00850"></a><span class="lineno"> 850</span> </div><div class="line"><a name="l00851"></a><span class="lineno"> 851</span>  <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>  <span class="comment">! beginning reconstruction of sum</span></div><div class="line"><a name="l00853"></a><span class="lineno"> 853</span>  <span class="comment">! from integer vector</span></div><div class="line"><a name="l00854"></a><span class="lineno"> 854</span> </div><div class="line"><a name="l00855"></a><span class="lineno"> 855</span>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_int)'</span></div><div class="line"><a name="l00856"></a><span class="lineno"> 856</span> </div><div class="line"><a name="l00857"></a><span class="lineno"> 857</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l00858"></a><span class="lineno"> 858</span> <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>  i8_radix = radix(ix_8)</div><div class="line"><a name="l00860"></a><span class="lineno"> 860</span> </div><div class="line"><a name="l00861"></a><span class="lineno"> 861</span> <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> <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>  <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l00864"></a><span class="lineno"> 864</span>  voffset = 3</div><div class="line"><a name="l00865"></a><span class="lineno"> 865</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00866"></a><span class="lineno"> 866</span>  voffset = 1</div><div class="line"><a name="l00867"></a><span class="lineno"> 867</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00868"></a><span class="lineno"> 868</span> </div><div class="line"><a name="l00869"></a><span class="lineno"> 869</span> <span class="comment">! compute offsets for each field</span></div><div class="line"><a name="l00870"></a><span class="lineno"> 870</span>  offset(1) = voffset</div><div class="line"><a name="l00871"></a><span class="lineno"> 871</span>  <span class="keywordflow">do</span> ifld=2,nflds</div><div class="line"><a name="l00872"></a><span class="lineno"> 872</span>  offset(ifld) = offset(ifld-1) &</div><div class="line"><a name="l00873"></a><span class="lineno"> 873</span>  + (max_levels(ifld-1) + voffset)</div><div class="line"><a name="l00874"></a><span class="lineno"> 874</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00875"></a><span class="lineno"> 875</span>  veclth = offset(nflds) + max_levels(nflds)</div><div class="line"><a name="l00876"></a><span class="lineno"> 876</span> </div><div class="line"><a name="l00877"></a><span class="lineno"> 877</span> <span class="comment">! split summand index range over OpenMP threads</span></div><div class="line"><a name="l00878"></a><span class="lineno"> 878</span>  <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> </div><div class="line"><a name="l00880"></a><span class="lineno"> 880</span> <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> <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> <span class="comment">! "normal" after level 1 calculation)</span></div><div class="line"><a name="l00883"></a><span class="lineno"> 883</span>  i8_arr_lsum_level(:) = 0_i8</div><div class="line"><a name="l00884"></a><span class="lineno"> 884</span> </div><div class="line"><a name="l00885"></a><span class="lineno"> 885</span> <span class="comment">!$omp parallel do &</span></div><div class="line"><a name="l00886"></a><span class="lineno"> 886</span> <span class="comment">!$omp default(shared) &</span></div><div class="line"><a name="l00887"></a><span class="lineno"> 887</span> <span class="comment">!$omp private(ithread, ifld, ioffset, isum, arr_frac, arr_exp, &</span></div><div class="line"><a name="l00888"></a><span class="lineno"> 888</span> <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>  <span class="keywordflow">do</span> ithread=1,omp_nthreads</div><div class="line"><a name="l00890"></a><span class="lineno"> 890</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">'repro_sum_loopb'</span>)</div><div class="line"><a name="l00891"></a><span class="lineno"> 891</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00892"></a><span class="lineno"> 892</span>  ioffset = offset(ifld)</div><div class="line"><a name="l00893"></a><span class="lineno"> 893</span> </div><div class="line"><a name="l00894"></a><span class="lineno"> 894</span>  max_error(ifld,ithread) = 0</div><div class="line"><a name="l00895"></a><span class="lineno"> 895</span>  not_exact(ifld,ithread) = 0</div><div class="line"><a name="l00896"></a><span class="lineno"> 896</span> </div><div class="line"><a name="l00897"></a><span class="lineno"> 897</span>  i8_arr_tlsum_level(:,ifld,ithread) = 0_i8</div><div class="line"><a name="l00898"></a><span class="lineno"> 898</span>  <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>  arr_remainder = 0.0_r8</div><div class="line"><a name="l00900"></a><span class="lineno"> 900</span> </div><div class="line"><a name="l00901"></a><span class="lineno"> 901</span>  <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>  arr_exp = exponent(arr(isum,ifld))</div><div class="line"><a name="l00903"></a><span class="lineno"> 903</span>  arr_frac = fraction(arr(isum,ifld))</div><div class="line"><a name="l00904"></a><span class="lineno"> 904</span> </div><div class="line"><a name="l00905"></a><span class="lineno"> 905</span> <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>  <span class="keywordflow">if</span> (arr_exp > arr_gmax_exp(ifld)) <span class="keywordflow">then</span></div><div class="line"><a name="l00907"></a><span class="lineno"> 907</span>  max_error(ifld,ithread) = 1</div><div class="line"><a name="l00908"></a><span class="lineno"> 908</span>  <span class="keywordflow">exit</span></div><div class="line"><a name="l00909"></a><span class="lineno"> 909</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00910"></a><span class="lineno"> 910</span> </div><div class="line"><a name="l00911"></a><span class="lineno"> 911</span> <span class="comment">! calculate first shift</span></div><div class="line"><a name="l00912"></a><span class="lineno"> 912</span>  arr_shift = arr_max_shift - (arr_gmax_exp(ifld)-arr_exp)</div><div class="line"><a name="l00913"></a><span class="lineno"> 913</span> </div><div class="line"><a name="l00914"></a><span class="lineno"> 914</span> <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> <span class="comment">! 'normal' - algorithm still works if this is not true)</span></div><div class="line"><a name="l00916"></a><span class="lineno"> 916</span> <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>  <span class="keywordflow">if</span> (arr_shift < 1) <span class="keywordflow">then</span></div><div class="line"><a name="l00918"></a><span class="lineno"> 918</span>  ilevel = (1 + (arr_gmax_exp(ifld)-arr_exp))/arr_max_shift</div><div class="line"><a name="l00919"></a><span class="lineno"> 919</span>  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> </div><div class="line"><a name="l00921"></a><span class="lineno"> 921</span>  <span class="keywordflow">do</span> <span class="keywordflow">while</span> (arr_shift < 1)</div><div class="line"><a name="l00922"></a><span class="lineno"> 922</span>  arr_shift = arr_shift + arr_max_shift</div><div class="line"><a name="l00923"></a><span class="lineno"> 923</span>  ilevel = ilevel + 1</div><div class="line"><a name="l00924"></a><span class="lineno"> 924</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00925"></a><span class="lineno"> 925</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l00926"></a><span class="lineno"> 926</span>  ilevel = 1</div><div class="line"><a name="l00927"></a><span class="lineno"> 927</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00928"></a><span class="lineno"> 928</span> </div><div class="line"><a name="l00929"></a><span class="lineno"> 929</span>  <span class="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> <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> <span class="comment">! sum, and calculate the remainder.</span></div><div class="line"><a name="l00932"></a><span class="lineno"> 932</span>  arr_remainder = scale(arr_frac,arr_shift)</div><div class="line"><a name="l00933"></a><span class="lineno"> 933</span>  i8_arr_level = int(arr_remainder,i8)</div><div class="line"><a name="l00934"></a><span class="lineno"> 934</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) = &</div><div class="line"><a name="l00935"></a><span class="lineno"> 935</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) + i8_arr_level</div><div class="line"><a name="l00936"></a><span class="lineno"> 936</span>  arr_remainder = arr_remainder - i8_arr_level</div><div class="line"><a name="l00937"></a><span class="lineno"> 937</span> </div><div class="line"><a name="l00938"></a><span class="lineno"> 938</span> <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> <span class="comment">! sum, and calculate new remainder</span></div><div class="line"><a name="l00940"></a><span class="lineno"> 940</span>  <span class="keywordflow">do</span> <span class="keywordflow">while</span> ((arr_remainder .ne. 0.0_r8) &</div><div class="line"><a name="l00941"></a><span class="lineno"> 941</span>  .and. (ilevel < max_levels(ifld)))</div><div class="line"><a name="l00942"></a><span class="lineno"> 942</span>  ilevel = ilevel + 1</div><div class="line"><a name="l00943"></a><span class="lineno"> 943</span>  arr_remainder = scale(arr_remainder,arr_max_shift)</div><div class="line"><a name="l00944"></a><span class="lineno"> 944</span>  i8_arr_level = int(arr_remainder,i8)</div><div class="line"><a name="l00945"></a><span class="lineno"> 945</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) = &</div><div class="line"><a name="l00946"></a><span class="lineno"> 946</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) + i8_arr_level</div><div class="line"><a name="l00947"></a><span class="lineno"> 947</span>  arr_remainder = arr_remainder - i8_arr_level</div><div class="line"><a name="l00948"></a><span class="lineno"> 948</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00949"></a><span class="lineno"> 949</span> </div><div class="line"><a name="l00950"></a><span class="lineno"> 950</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00951"></a><span class="lineno"> 951</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00952"></a><span class="lineno"> 952</span> </div><div class="line"><a name="l00953"></a><span class="lineno"> 953</span>  <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>  not_exact(ifld,ithread) = 1</div><div class="line"><a name="l00955"></a><span class="lineno"> 955</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00956"></a><span class="lineno"> 956</span> </div><div class="line"><a name="l00957"></a><span class="lineno"> 957</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00958"></a><span class="lineno"> 958</span> </div><div class="line"><a name="l00959"></a><span class="lineno"> 959</span> <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> <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> <span class="comment">! (radix(IX_8)**arr_max_shift), add this 'overlap' to next larger integer in </span></div><div class="line"><a name="l00962"></a><span class="lineno"> 962</span> <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> <span class="comment">! "ilevel-1==0" 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> <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>  <span class="keywordflow">do</span> ilevel=max_levels(ifld),1,-1</div><div class="line"><a name="l00966"></a><span class="lineno"> 966</span>  rx_8 = i8_arr_tlsum_level(ilevel,ifld,ithread)</div><div class="line"><a name="l00967"></a><span class="lineno"> 967</span>  ix_8 = int(scale(rx_8,-arr_max_shift),i8)</div><div class="line"><a name="l00968"></a><span class="lineno"> 968</span>  <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>  i8_arr_tlsum_level(ilevel-1,ifld,ithread) = &</div><div class="line"><a name="l00970"></a><span class="lineno"> 970</span>  i8_arr_tlsum_level(ilevel-1,ifld,ithread) + ix_8</div><div class="line"><a name="l00971"></a><span class="lineno"> 971</span>  ix_8 = ix_8*(i8_radix**arr_max_shift)</div><div class="line"><a name="l00972"></a><span class="lineno"> 972</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) = &</div><div class="line"><a name="l00973"></a><span class="lineno"> 973</span>  i8_arr_tlsum_level(ilevel,ifld,ithread) - ix_8</div><div class="line"><a name="l00974"></a><span class="lineno"> 974</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l00975"></a><span class="lineno"> 975</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00976"></a><span class="lineno"> 976</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00977"></a><span class="lineno"> 977</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">'repro_sum_loopb'</span>)</div><div class="line"><a name="l00978"></a><span class="lineno"> 978</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00979"></a><span class="lineno"> 979</span> </div><div class="line"><a name="l00980"></a><span class="lineno"> 980</span> <span class="comment">! sum contributions from different threads</span></div><div class="line"><a name="l00981"></a><span class="lineno"> 981</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00982"></a><span class="lineno"> 982</span>  ioffset = offset(ifld)</div><div class="line"><a name="l00983"></a><span class="lineno"> 983</span>  <span class="keywordflow">do</span> ithread = 1,omp_nthreads</div><div class="line"><a name="l00984"></a><span class="lineno"> 984</span>  <span class="keywordflow">do</span> ilevel = 0,max_levels(ifld)</div><div class="line"><a name="l00985"></a><span class="lineno"> 985</span>  i8_arr_lsum_level(ioffset+ilevel) = &</div><div class="line"><a name="l00986"></a><span class="lineno"> 986</span>  i8_arr_lsum_level(ioffset+ilevel) &</div><div class="line"><a name="l00987"></a><span class="lineno"> 987</span>  + i8_arr_tlsum_level(ilevel,ifld,ithread)</div><div class="line"><a name="l00988"></a><span class="lineno"> 988</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00989"></a><span class="lineno"> 989</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00990"></a><span class="lineno"> 990</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l00991"></a><span class="lineno"> 991</span> </div><div class="line"><a name="l00992"></a><span class="lineno"> 992</span> <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> <span class="comment">! before full accuracy was achieved</span></div><div class="line"><a name="l00994"></a><span class="lineno"> 994</span>  <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l00995"></a><span class="lineno"> 995</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l00996"></a><span class="lineno"> 996</span>  ioffset = offset(ifld)</div><div class="line"><a name="l00997"></a><span class="lineno"> 997</span>  i8_arr_lsum_level(ioffset-voffset+1) = maxval(max_error(ifld,:))</div><div class="line"><a name="l00998"></a><span class="lineno"> 998</span>  i8_arr_lsum_level(ioffset-voffset+2) = maxval(not_exact(ifld,:))</div><div class="line"><a name="l00999"></a><span class="lineno"> 999</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01000"></a><span class="lineno"> 1000</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01001"></a><span class="lineno"> 1001</span> </div><div class="line"><a name="l01002"></a><span class="lineno"> 1002</span> <span class="comment">! sum integer vector element-wise</span></div><div class="line"><a name="l01003"></a><span class="lineno"> 1003</span> <span class="preprocessor">#if ( defined noI8 )</span></div><div class="line"><a name="l01004"></a><span class="lineno"> 1004</span> <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>  <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">"repro_sum_allr_i4"</span>)</div><div class="line"><a name="l01006"></a><span class="lineno"> 1006</span>  <span class="keyword">call </span>mpi_allreduce (i8_arr_lsum_level, i8_arr_gsum_level, & </div><div class="line"><a name="l01007"></a><span class="lineno"> 1007</span>  veclth, mpi_integer, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l01008"></a><span class="lineno"> 1008</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">"repro_sum_allr_i4"</span>)</div><div class="line"><a name="l01009"></a><span class="lineno"> 1009</span> <span class="preprocessor">#else</span></div><div class="line"><a name="l01010"></a><span class="lineno"> 1010</span> <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">"repro_sum_allr_i8"</span>)</div><div class="line"><a name="l01011"></a><span class="lineno"> 1011</span>  <span class="keyword">call </span>mpi_allreduce (i8_arr_lsum_level, i8_arr_gsum_level, & </div><div class="line"><a name="l01012"></a><span class="lineno"> 1012</span>  veclth, mpi_integer8, mpi_sum, mpi_comm, ierr)</div><div class="line"><a name="l01013"></a><span class="lineno"> 1013</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#a01a76142a72cc6bc845736e79ad99bd3">oasis_timer_stop</a>(<span class="stringliteral">"repro_sum_allr_i8"</span>)</div><div class="line"><a name="l01014"></a><span class="lineno"> 1014</span> <span class="preprocessor">#endif</span></div><div class="line"><a name="l01015"></a><span class="lineno"> 1015</span> <span class="preprocessor"></span></div><div class="line"><a name="l01016"></a><span class="lineno"> 1016</span> <span class="comment">! Construct global sum from integer vector representation:</span></div><div class="line"><a name="l01017"></a><span class="lineno"> 1017</span> <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> <span class="comment">! When shifting back, need to "add back in" true arr_gmax exponent. This was</span></div><div class="line"><a name="l01019"></a><span class="lineno"> 1019</span> <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> <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> <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> <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> <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> <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> <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> <span class="comment">! size is greater than the max number of 'digits' in r8 value (if xmax_nsummands </span></div><div class="line"><a name="l01027"></a><span class="lineno"> 1027</span> <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> <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> <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> <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> <span class="comment">! max_jlevel parameter calculation is correct).</span></div><div class="line"><a name="l01032"></a><span class="lineno"> 1032</span> </div><div class="line"><a name="l01033"></a><span class="lineno"> 1033</span>  recompute = .false.</div><div class="line"><a name="l01034"></a><span class="lineno"> 1034</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01035"></a><span class="lineno"> 1035</span>  arr_gsum(ifld) = 0.0_r8</div><div class="line"><a name="l01036"></a><span class="lineno"> 1036</span>  ioffset = offset(ifld)</div><div class="line"><a name="l01037"></a><span class="lineno"> 1037</span> </div><div class="line"><a name="l01038"></a><span class="lineno"> 1038</span> <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> <span class="comment">! was exceeded on any of the processes</span></div><div class="line"><a name="l01040"></a><span class="lineno"> 1040</span>  <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l01041"></a><span class="lineno"> 1041</span>  <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>  recompute = .true.</div><div class="line"><a name="l01043"></a><span class="lineno"> 1043</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01044"></a><span class="lineno"> 1044</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01045"></a><span class="lineno"> 1045</span> </div><div class="line"><a name="l01046"></a><span class="lineno"> 1046</span>  <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> </div><div class="line"><a name="l01048"></a><span class="lineno"> 1048</span> <span class="comment">! preprocess integer vector:</span></div><div class="line"><a name="l01049"></a><span class="lineno"> 1049</span> <span class="comment">! a) if value larger than or equal to (radix(IX_8)**arr_max_shift), add this 'overlap' </span></div><div class="line"><a name="l01050"></a><span class="lineno"> 1050</span> <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> <span class="comment">! component. Note that have "ilevel-1=0" level here as described above.</span></div><div class="line"><a name="l01052"></a><span class="lineno"> 1052</span>  <span class="keywordflow">do</span> ilevel=max_levels(ifld),1,-1</div><div class="line"><a name="l01053"></a><span class="lineno"> 1053</span>  rx_8 = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01054"></a><span class="lineno"> 1054</span>  ix_8 = int(scale(rx_8,-arr_max_shift),i8)</div><div class="line"><a name="l01055"></a><span class="lineno"> 1055</span>  <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>  i8_arr_gsum_level(ioffset+ilevel-1) = i8_arr_gsum_level(ioffset+ilevel-1) &</div><div class="line"><a name="l01057"></a><span class="lineno"> 1057</span>  + ix_8</div><div class="line"><a name="l01058"></a><span class="lineno"> 1058</span>  ix_8 = ix_8*(i8_radix**arr_max_shift)</div><div class="line"><a name="l01059"></a><span class="lineno"> 1059</span>  i8_arr_gsum_level(ioffset+ilevel) = i8_arr_gsum_level(ioffset+ilevel) &</div><div class="line"><a name="l01060"></a><span class="lineno"> 1060</span>  - ix_8</div><div class="line"><a name="l01061"></a><span class="lineno"> 1061</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01062"></a><span class="lineno"> 1062</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01063"></a><span class="lineno"> 1063</span> <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> <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> <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> <span class="comment">! reconstructing r8 sum from integer vector)</span></div><div class="line"><a name="l01067"></a><span class="lineno"> 1067</span>  ilevel = 0</div><div class="line"><a name="l01068"></a><span class="lineno"> 1068</span>  <span class="keywordflow">do</span> <span class="keywordflow">while</span> ((i8_arr_gsum_level(ioffset+ilevel) .eq. 0_i8) &</div><div class="line"><a name="l01069"></a><span class="lineno"> 1069</span>  .and. (ilevel < max_levels(ifld)))</div><div class="line"><a name="l01070"></a><span class="lineno"> 1070</span>  ilevel = ilevel + 1</div><div class="line"><a name="l01071"></a><span class="lineno"> 1071</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01072"></a><span class="lineno"> 1072</span> <span class="comment">!</span></div><div class="line"><a name="l01073"></a><span class="lineno"> 1073</span>  <span class="keywordflow">if</span> (ilevel < max_levels(ifld)) <span class="keywordflow">then</span></div><div class="line"><a name="l01074"></a><span class="lineno"> 1074</span>  <span class="keywordflow">if</span> (i8_arr_gsum_level(ioffset+ilevel) > 0_i8) <span class="keywordflow">then</span></div><div class="line"><a name="l01075"></a><span class="lineno"> 1075</span>  i8_sign = 1_i8</div><div class="line"><a name="l01076"></a><span class="lineno"> 1076</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01077"></a><span class="lineno"> 1077</span>  i8_sign = -1_i8</div><div class="line"><a name="l01078"></a><span class="lineno"> 1078</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01079"></a><span class="lineno"> 1079</span>  <span class="keywordflow">do</span> jlevel=ilevel,max_levels(ifld)-1</div><div class="line"><a name="l01080"></a><span class="lineno"> 1080</span>  <span class="keywordflow">if</span> (sign(1_i8,i8_arr_gsum_level(ioffset+jlevel)) &</div><div class="line"><a name="l01081"></a><span class="lineno"> 1081</span>  .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>  i8_arr_gsum_level(ioffset+jlevel) = i8_arr_gsum_level(ioffset+jlevel) &</div><div class="line"><a name="l01083"></a><span class="lineno"> 1083</span>  - i8_sign</div><div class="line"><a name="l01084"></a><span class="lineno"> 1084</span>  i8_arr_gsum_level(ioffset+jlevel+1) = i8_arr_gsum_level(ioffset+jlevel+1) &</div><div class="line"><a name="l01085"></a><span class="lineno"> 1085</span>  + i8_sign*(i8_radix**arr_max_shift)</div><div class="line"><a name="l01086"></a><span class="lineno"> 1086</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01087"></a><span class="lineno"> 1087</span> <span class="keywordflow"> enddo</span> </div><div class="line"><a name="l01088"></a><span class="lineno"> 1088</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01089"></a><span class="lineno"> 1089</span> </div><div class="line"><a name="l01090"></a><span class="lineno"> 1090</span> <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>  arr_shift = arr_gmax_exp(ifld) &</div><div class="line"><a name="l01092"></a><span class="lineno"> 1092</span>  - max_levels(ifld)*arr_max_shift</div><div class="line"><a name="l01093"></a><span class="lineno"> 1093</span>  curr_exp = 0</div><div class="line"><a name="l01094"></a><span class="lineno"> 1094</span>  first = .true.</div><div class="line"><a name="l01095"></a><span class="lineno"> 1095</span>  <span class="keywordflow">do</span> ilevel=max_levels(ifld),0,-1</div><div class="line"><a name="l01096"></a><span class="lineno"> 1096</span> </div><div class="line"><a name="l01097"></a><span class="lineno"> 1097</span>  <span class="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>  jlevel = 1</div><div class="line"><a name="l01099"></a><span class="lineno"> 1099</span> </div><div class="line"><a name="l01100"></a><span class="lineno"> 1100</span> <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>  x_8(jlevel) = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01102"></a><span class="lineno"> 1102</span>  lx(jlevel) = exponent(x_8(jlevel))</div><div class="line"><a name="l01103"></a><span class="lineno"> 1103</span> </div><div class="line"><a name="l01104"></a><span class="lineno"> 1104</span> <span class="comment">! calculate remainder</span></div><div class="line"><a name="l01105"></a><span class="lineno"> 1105</span>  ix_8 = int(x_8(jlevel),i8)</div><div class="line"><a name="l01106"></a><span class="lineno"> 1106</span>  rx_8 = (i8_arr_gsum_level(ioffset+ilevel) - ix_8)</div><div class="line"><a name="l01107"></a><span class="lineno"> 1107</span> </div><div class="line"><a name="l01108"></a><span class="lineno"> 1108</span> <span class="comment">! repeat using remainder</span></div><div class="line"><a name="l01109"></a><span class="lineno"> 1109</span>  <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>  jlevel = jlevel + 1</div><div class="line"><a name="l01111"></a><span class="lineno"> 1111</span>  x_8(jlevel) = rx_8</div><div class="line"><a name="l01112"></a><span class="lineno"> 1112</span>  lx(jlevel) = exponent(rx_8)</div><div class="line"><a name="l01113"></a><span class="lineno"> 1113</span>  ix_8 = ix_8 + int(rx_8,i8)</div><div class="line"><a name="l01114"></a><span class="lineno"> 1114</span>  rx_8 = (i8_arr_gsum_level(ioffset+ilevel) - ix_8)</div><div class="line"><a name="l01115"></a><span class="lineno"> 1115</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01116"></a><span class="lineno"> 1116</span> </div><div class="line"><a name="l01117"></a><span class="lineno"> 1117</span> <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> <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> <span class="comment">! larger than minexponent</span></div><div class="line"><a name="l01120"></a><span class="lineno"> 1120</span>  <span class="keywordflow">do</span> <span class="keywordflow">while</span> (jlevel > 0)</div><div class="line"><a name="l01121"></a><span class="lineno"> 1121</span>  <span class="keywordflow">if</span> (first) <span class="keywordflow">then</span></div><div class="line"><a name="l01122"></a><span class="lineno"> 1122</span>  curr_exp = lx(jlevel) + arr_shift</div><div class="line"><a name="l01123"></a><span class="lineno"> 1123</span>  arr_gsum(ifld) = fraction(x_8(jlevel))</div><div class="line"><a name="l01124"></a><span class="lineno"> 1124</span>  first = .false.</div><div class="line"><a name="l01125"></a><span class="lineno"> 1125</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01126"></a><span class="lineno"> 1126</span>  corr_exp = curr_exp - (lx(jlevel) + arr_shift)</div><div class="line"><a name="l01127"></a><span class="lineno"> 1127</span>  arr_gsum(ifld) = fraction(x_8(jlevel)) &</div><div class="line"><a name="l01128"></a><span class="lineno"> 1128</span>  + scale(arr_gsum(ifld),corr_exp)</div><div class="line"><a name="l01129"></a><span class="lineno"> 1129</span>  curr_exp = lx(jlevel) + arr_shift</div><div class="line"><a name="l01130"></a><span class="lineno"> 1130</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01131"></a><span class="lineno"> 1131</span>  jlevel = jlevel - 1</div><div class="line"><a name="l01132"></a><span class="lineno"> 1132</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01133"></a><span class="lineno"> 1133</span> </div><div class="line"><a name="l01134"></a><span class="lineno"> 1134</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01135"></a><span class="lineno"> 1135</span> </div><div class="line"><a name="l01136"></a><span class="lineno"> 1136</span>  arr_shift = arr_shift + arr_max_shift</div><div class="line"><a name="l01137"></a><span class="lineno"> 1137</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01138"></a><span class="lineno"> 1138</span> </div><div class="line"><a name="l01139"></a><span class="lineno"> 1139</span> <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> <span class="comment">! to apply directly</span></div><div class="line"><a name="l01141"></a><span class="lineno"> 1141</span>  corr_exp = curr_exp + exponent(arr_gsum(ifld))</div><div class="line"><a name="l01142"></a><span class="lineno"> 1142</span>  <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>  arr_gsum(ifld) = set_exponent(arr_gsum(ifld),corr_exp)</div><div class="line"><a name="l01144"></a><span class="lineno"> 1144</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01145"></a><span class="lineno"> 1145</span>  rx_8 = set_exponent(arr_gsum(ifld), &</div><div class="line"><a name="l01146"></a><span class="lineno"> 1146</span>  corr_exp-minexponent(1._r8))</div><div class="line"><a name="l01147"></a><span class="lineno"> 1147</span>  arr_gsum(ifld) = scale(rx_8,minexponent(1._r8))</div><div class="line"><a name="l01148"></a><span class="lineno"> 1148</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01149"></a><span class="lineno"> 1149</span> </div><div class="line"><a name="l01150"></a><span class="lineno"> 1150</span> <span class="comment">! if validate is .true. and some precision lost, test whether 'too much' </span></div><div class="line"><a name="l01151"></a><span class="lineno"> 1151</span> <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> <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> <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>  <span class="keywordflow">if</span> (validate) <span class="keywordflow">then</span></div><div class="line"><a name="l01155"></a><span class="lineno"> 1155</span>  <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> </div><div class="line"><a name="l01157"></a><span class="lineno"> 1157</span> <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> <span class="comment">! subsequent levels contribute arr_max_shift digits.</span></div><div class="line"><a name="l01159"></a><span class="lineno"> 1159</span>  sum_digits = 0</div><div class="line"><a name="l01160"></a><span class="lineno"> 1160</span>  <span class="keywordflow">do</span> ilevel=0,max_levels(ifld)</div><div class="line"><a name="l01161"></a><span class="lineno"> 1161</span>  <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>  <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>  x_8(1) = i8_arr_gsum_level(ioffset+ilevel)</div><div class="line"><a name="l01164"></a><span class="lineno"> 1164</span>  lx(1) = exponent(x_8(1))</div><div class="line"><a name="l01165"></a><span class="lineno"> 1165</span>  sum_digits = lx(1)</div><div class="line"><a name="l01166"></a><span class="lineno"> 1166</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01167"></a><span class="lineno"> 1167</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01168"></a><span class="lineno"> 1168</span>  sum_digits = sum_digits + arr_max_shift</div><div class="line"><a name="l01169"></a><span class="lineno"> 1169</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01170"></a><span class="lineno"> 1170</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01171"></a><span class="lineno"> 1171</span> </div><div class="line"><a name="l01172"></a><span class="lineno"> 1172</span>  <span class="keywordflow">if</span> (sum_digits < digits(1.0_r8)) <span class="keywordflow">then</span></div><div class="line"><a name="l01173"></a><span class="lineno"> 1173</span>  recompute = .true.</div><div class="line"><a name="l01174"></a><span class="lineno"> 1174</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01175"></a><span class="lineno"> 1175</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01176"></a><span class="lineno"> 1176</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01177"></a><span class="lineno"> 1177</span> </div><div class="line"><a name="l01178"></a><span class="lineno"> 1178</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01179"></a><span class="lineno"> 1179</span> </div><div class="line"><a name="l01180"></a><span class="lineno"> 1180</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01181"></a><span class="lineno"> 1181</span> </div><div class="line"><a name="l01182"></a><span class="lineno"> 1182</span> </div><div class="line"><a name="l01183"></a><span class="lineno"> 1183</span> <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> </div><div class="line"><a name="l01185"></a><span class="lineno"> 1185</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l01186"></a><span class="lineno"> 1186</span> <span class="comment">! </span></div><div class="line"><a name="l01187"></a><span class="lineno"> 1187</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l01188"></a><span class="lineno"> 1188</span> <span class="comment">!> Test whether distributed sum exceeds tolerance and print out a</span></div><div class="line"><a name="l01189"></a><span class="lineno"> 1189</span> <span class="comment">!> warning message.</span></div><div class="line"><a name="l01190"></a><span class="lineno"> 1190</span> <span class="comment">! </span></div><div class="line"><a name="l01191"></a><span class="lineno"> 1191</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01192"></a><span class="lineno"> 1192</span> </div><div class="line"><a name="l01193"></a><span class="lineno"> 1193</span> <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, &</div><div class="line"><a name="l01194"></a><span class="lineno"> 1194</span>  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> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01196"></a><span class="lineno"> 1196</span> </div><div class="line"><a name="l01197"></a><span class="lineno"> 1197</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l01198"></a><span class="lineno"> 1198</span> </div><div class="line"><a name="l01199"></a><span class="lineno"> 1199</span>  <span class="keywordtype">character(len=*)</span>, <span class="keywordtype">intent(in)</span> :: name<span class="comment"> !< distributed sum identifier</span></div><div class="line"><a name="l01200"></a><span class="lineno"> 1200</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nflds<span class="comment"> !< number of fields</span></div><div class="line"><a name="l01201"></a><span class="lineno"> 1201</span>  <span class="keywordtype">logical</span>, <span class="keywordtype">intent(in)</span> :: master<span class="comment"> !< process that will write</span></div><div class="line"><a name="l01202"></a><span class="lineno"> 1202</span> <span class="comment"> !< warning messages?</span></div><div class="line"><a name="l01203"></a><span class="lineno"> 1203</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">optional</span>, <span class="keywordtype">intent(in)</span> :: logunit<span class="comment">!< unit warning messages </span></div><div class="line"><a name="l01204"></a><span class="lineno"> 1204</span> <span class="comment"> !< written to</span></div><div class="line"><a name="l01205"></a><span class="lineno"> 1205</span>  <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>  <span class="comment">!< relative and absolute</span></div><div class="line"><a name="l01207"></a><span class="lineno"> 1207</span> <span class="comment"> !< differences between fixed </span></div><div class="line"><a name="l01208"></a><span class="lineno"> 1208</span> <span class="comment"> !< and floating point sums</span></div><div class="line"><a name="l01209"></a><span class="lineno"> 1209</span> </div><div class="line"><a name="l01210"></a><span class="lineno"> 1210</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l01211"></a><span class="lineno"> 1211</span> </div><div class="line"><a name="l01212"></a><span class="lineno"> 1212</span>  <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>  <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>  <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>  <span class="comment">! sum exceeds tolerance</span></div><div class="line"><a name="l01216"></a><span class="lineno"> 1216</span>  <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>  <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>  <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>  <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>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_tolExceeded)'</span></div><div class="line"><a name="l01221"></a><span class="lineno"> 1221</span> </div><div class="line"><a name="l01222"></a><span class="lineno"> 1222</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01223"></a><span class="lineno"> 1223</span> </div><div class="line"><a name="l01224"></a><span class="lineno"> 1224</span>  <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>  <span class="keywordflow">if</span> (<a class="code" href="namespacemod__oasis__reprosum.html#a6a80f4226d35d0842617e6eb6c9406e1">oasis_reprosum_reldiffmax</a> < 0.0_r8) <span class="keywordflow">return</span></div><div class="line"><a name="l01226"></a><span class="lineno"> 1226</span> </div><div class="line"><a name="l01227"></a><span class="lineno"> 1227</span>  <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>  llogunit = logunit</div><div class="line"><a name="l01229"></a><span class="lineno"> 1229</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01230"></a><span class="lineno"> 1230</span>  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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01232"></a><span class="lineno"> 1232</span> </div><div class="line"><a name="l01233"></a><span class="lineno"> 1233</span>  <span class="comment">! check that "fast" reproducible sum is accurate enough.</span></div><div class="line"><a name="l01234"></a><span class="lineno"> 1234</span>  exceeds_limit = 0</div><div class="line"><a name="l01235"></a><span class="lineno"> 1235</span>  max_rel_diff = 0.0_r8</div><div class="line"><a name="l01236"></a><span class="lineno"> 1236</span>  max_abs_diff = 0.0_r8</div><div class="line"><a name="l01237"></a><span class="lineno"> 1237</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01238"></a><span class="lineno"> 1238</span>  <span class="keywordflow">if</span> (rel_diff(1,ifld) > <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>  exceeds_limit = exceeds_limit + 1</div><div class="line"><a name="l01240"></a><span class="lineno"> 1240</span>  <span class="keywordflow">if</span> (rel_diff(1,ifld) > max_rel_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l01241"></a><span class="lineno"> 1241</span>  max_rel_diff = rel_diff(1,ifld)</div><div class="line"><a name="l01242"></a><span class="lineno"> 1242</span>  max_rel_diff_idx = ifld</div><div class="line"><a name="l01243"></a><span class="lineno"> 1243</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01244"></a><span class="lineno"> 1244</span>  <span class="keywordflow">if</span> (rel_diff(2,ifld) > max_abs_diff) <span class="keywordflow">then</span></div><div class="line"><a name="l01245"></a><span class="lineno"> 1245</span>  max_abs_diff = rel_diff(2,ifld)</div><div class="line"><a name="l01246"></a><span class="lineno"> 1246</span>  max_abs_diff_idx = ifld</div><div class="line"><a name="l01247"></a><span class="lineno"> 1247</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01248"></a><span class="lineno"> 1248</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01249"></a><span class="lineno"> 1249</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01250"></a><span class="lineno"> 1250</span> </div><div class="line"><a name="l01251"></a><span class="lineno"> 1251</span>  <span class="keywordflow">if</span> (exceeds_limit > 0) <span class="keywordflow">then</span></div><div class="line"><a name="l01252"></a><span class="lineno"> 1252</span>  <span class="keywordflow">if</span> (master) <span class="keywordflow">then</span></div><div class="line"><a name="l01253"></a><span class="lineno"> 1253</span>  <span class="keyword">write</span>(llogunit,*) subname,trim(name), &</div><div class="line"><a name="l01254"></a><span class="lineno"> 1254</span>  <span class="stringliteral">': difference in fixed and floating point sums '</span>, &</div><div class="line"><a name="l01255"></a><span class="lineno"> 1255</span>  <span class="stringliteral">' exceeds tolerance in '</span>, exceeds_limit, &</div><div class="line"><a name="l01256"></a><span class="lineno"> 1256</span>  <span class="stringliteral">' fields.'</span></div><div class="line"><a name="l01257"></a><span class="lineno"> 1257</span>  <span class="keyword">write</span>(llogunit,*) subname,<span class="stringliteral">' Maximum relative diff: (rel)'</span>, &</div><div class="line"><a name="l01258"></a><span class="lineno"> 1258</span>  rel_diff(1,max_rel_diff_idx), <span class="stringliteral">' (abs) '</span>, &</div><div class="line"><a name="l01259"></a><span class="lineno"> 1259</span>  rel_diff(2,max_rel_diff_idx)</div><div class="line"><a name="l01260"></a><span class="lineno"> 1260</span>  <span class="keyword">write</span>(llogunit,*) subname,<span class="stringliteral">' Maximum absolute diff: (rel)'</span>, &</div><div class="line"><a name="l01261"></a><span class="lineno"> 1261</span>  rel_diff(1,max_abs_diff_idx), <span class="stringliteral">' (abs) '</span>, &</div><div class="line"><a name="l01262"></a><span class="lineno"> 1262</span>  rel_diff(2,max_abs_diff_idx)</div><div class="line"><a name="l01263"></a><span class="lineno"> 1263</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01264"></a><span class="lineno"> 1264</span>  <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> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01266"></a><span class="lineno"> 1266</span> </div><div class="line"><a name="l01267"></a><span class="lineno"> 1267</span> </div><div class="line"><a name="l01268"></a><span class="lineno"> 1268</span> <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> </div><div class="line"><a name="l01270"></a><span class="lineno"> 1270</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l01271"></a><span class="lineno"> 1271</span> <span class="comment">! </span></div><div class="line"><a name="l01272"></a><span class="lineno"> 1272</span> <span class="comment">! Purpose: </span><span class="comment"></span></div><div class="line"><a name="l01273"></a><span class="lineno"> 1273</span> <span class="comment">!> Compute the global sum of each field in "arr" using the indicated </span></div><div class="line"><a name="l01274"></a><span class="lineno"> 1274</span> <span class="comment">!> communicator with a reproducible yet scalable implementation based </span></div><div class="line"><a name="l01275"></a><span class="lineno"> 1275</span> <span class="comment">!> on He and Ding's implementation of the double-double algorithm.</span></div><div class="line"><a name="l01276"></a><span class="lineno"> 1276</span> <span class="comment">! </span></div><div class="line"><a name="l01277"></a><span class="lineno"> 1277</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01278"></a><span class="lineno"> 1278</span> </div><div class="line"><a name="l01279"></a><span class="lineno"> 1279</span> <span class="keyword"> subroutine </span><a class="code" href="namespacemod__oasis__reprosum.html#afe237975bf8e718f32523c3929911102">oasis_reprosum_ddpdd</a> (arr, arr_gsum, nsummands, dsummands, &</div><div class="line"><a name="l01280"></a><span class="lineno"> 1280</span>  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> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01282"></a><span class="lineno"> 1282</span> </div><div class="line"><a name="l01283"></a><span class="lineno"> 1283</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l01284"></a><span class="lineno"> 1284</span> </div><div class="line"><a name="l01285"></a><span class="lineno"> 1285</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nsummands<span class="comment"> !< number of local summands</span></div><div class="line"><a name="l01286"></a><span class="lineno"> 1286</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: dsummands<span class="comment"> !< declared first dimension</span></div><div class="line"><a name="l01287"></a><span class="lineno"> 1287</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: nflds<span class="comment"> !< number of fields</span></div><div class="line"><a name="l01288"></a><span class="lineno"> 1288</span>  <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>  <span class="comment">!< input array</span></div><div class="line"><a name="l01290"></a><span class="lineno"> 1290</span>  <span class="keywordtype">integer</span>, <span class="keywordtype">intent(in)</span> :: mpi_comm<span class="comment"> !< MPI subcommunicator</span></div><div class="line"><a name="l01291"></a><span class="lineno"> 1291</span> </div><div class="line"><a name="l01292"></a><span class="lineno"> 1292</span>  <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>  <span class="comment">!< global sums</span></div><div class="line"><a name="l01294"></a><span class="lineno"> 1294</span> </div><div class="line"><a name="l01295"></a><span class="lineno"> 1295</span> </div><div class="line"><a name="l01296"></a><span class="lineno"> 1296</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l01297"></a><span class="lineno"> 1297</span> </div><div class="line"><a name="l01298"></a><span class="lineno"> 1298</span>  <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>  <span class="comment">! current arithmetic mode</span></div><div class="line"><a name="l01300"></a><span class="lineno"> 1300</span>  <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>  <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> </div><div class="line"><a name="l01303"></a><span class="lineno"> 1303</span>  <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>  <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>  <span class="comment">! format)</span></div><div class="line"><a name="l01306"></a><span class="lineno"> 1306</span>  <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>  <span class="comment">! format)</span></div><div class="line"><a name="l01308"></a><span class="lineno"> 1308</span> </div><div class="line"><a name="l01309"></a><span class="lineno"> 1309</span>  <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>  <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>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_ddpdd)'</span></div><div class="line"><a name="l01312"></a><span class="lineno"> 1312</span> </div><div class="line"><a name="l01313"></a><span class="lineno"> 1313</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01314"></a><span class="lineno"> 1314</span> </div><div class="line"><a name="l01315"></a><span class="lineno"> 1315</span>  <span class="keyword">call </span>shr_reprosumx86_fix_start (old_cw)</div><div class="line"><a name="l01316"></a><span class="lineno"> 1316</span> </div><div class="line"><a name="l01317"></a><span class="lineno"> 1317</span>  <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>  <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>  first_time = .false.</div><div class="line"><a name="l01320"></a><span class="lineno"> 1320</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01321"></a><span class="lineno"> 1321</span> </div><div class="line"><a name="l01322"></a><span class="lineno"> 1322</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01323"></a><span class="lineno"> 1323</span>  arr_lsum_dd(ifld) = (0.0_r8,0.0_r8)</div><div class="line"><a name="l01324"></a><span class="lineno"> 1324</span> </div><div class="line"><a name="l01325"></a><span class="lineno"> 1325</span>  <span class="keywordflow">do</span> isum=1,nsummands</div><div class="line"><a name="l01326"></a><span class="lineno"> 1326</span>  </div><div class="line"><a name="l01327"></a><span class="lineno"> 1327</span>  <span class="comment">! Compute arr(isum,ifld) + arr_lsum_dd(ifld) using Knuth''s </span></div><div class="line"><a name="l01328"></a><span class="lineno"> 1328</span>  <span class="comment">! trick.</span></div><div class="line"><a name="l01329"></a><span class="lineno"> 1329</span>  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>  e = t1 - arr(isum,ifld)</div><div class="line"><a name="l01331"></a><span class="lineno"> 1331</span>  t2 = ((<span class="keywordtype">real(arr_lsum_dd(ifld))</span> - e) &</div><div class="line"><a name="l01332"></a><span class="lineno"> 1332</span>  + (arr(isum,ifld) - (t1 - e))) &</div><div class="line"><a name="l01333"></a><span class="lineno"> 1333</span>  + aimag(arr_lsum_dd(ifld))</div><div class="line"><a name="l01334"></a><span class="lineno"> 1334</span> </div><div class="line"><a name="l01335"></a><span class="lineno"> 1335</span>  <span class="comment">! The result is t1 + t2, after normalization.</span></div><div class="line"><a name="l01336"></a><span class="lineno"> 1336</span>  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> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01338"></a><span class="lineno"> 1338</span> </div><div class="line"><a name="l01339"></a><span class="lineno"> 1339</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01340"></a><span class="lineno"> 1340</span> </div><div class="line"><a name="l01341"></a><span class="lineno"> 1341</span>  <span class="keyword">call </span>mpi_allreduce (arr_lsum_dd, arr_gsum_dd, nflds, &</div><div class="line"><a name="l01342"></a><span class="lineno"> 1342</span>  mpi_complex16, mpi_sumdd, mpi_comm, ierr)</div><div class="line"><a name="l01343"></a><span class="lineno"> 1343</span>  <span class="keywordflow">do</span> ifld=1,nflds</div><div class="line"><a name="l01344"></a><span class="lineno"> 1344</span>  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> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01346"></a><span class="lineno"> 1346</span> </div><div class="line"><a name="l01347"></a><span class="lineno"> 1347</span>  <span class="keyword">call </span>shr_reprosumx86_fix_end (old_cw)</div><div class="line"><a name="l01348"></a><span class="lineno"> 1348</span> </div><div class="line"><a name="l01349"></a><span class="lineno"> 1349</span> <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> </div><div class="line"><a name="l01351"></a><span class="lineno"> 1351</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01352"></a><span class="lineno"> 1352</span> </div><div class="line"><a name="l01353"></a><span class="lineno"> 1353</span> <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> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01355"></a><span class="lineno"> 1355</span> <span class="comment">! </span></div><div class="line"><a name="l01356"></a><span class="lineno"> 1356</span> <span class="comment">! Purpose: </span></div><div class="line"><a name="l01357"></a><span class="lineno"> 1357</span> <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> <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> <span class="comment">! </span></div><div class="line"><a name="l01360"></a><span class="lineno"> 1360</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01361"></a><span class="lineno"> 1361</span> </div><div class="line"><a name="l01362"></a><span class="lineno"> 1362</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l01363"></a><span class="lineno"> 1363</span> </div><div class="line"><a name="l01364"></a><span class="lineno"> 1364</span>  <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>  <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>  <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>  <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> </div><div class="line"><a name="l01369"></a><span class="lineno"> 1369</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l01370"></a><span class="lineno"> 1370</span> </div><div class="line"><a name="l01371"></a><span class="lineno"> 1371</span>  <span class="keywordtype">real(r8)</span> e, t1, t2</div><div class="line"><a name="l01372"></a><span class="lineno"> 1372</span>  <span class="keywordtype">integer</span> i</div><div class="line"><a name="l01373"></a><span class="lineno"> 1373</span>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_mod:DDPDD)'</span></div><div class="line"><a name="l01374"></a><span class="lineno"> 1374</span> </div><div class="line"><a name="l01375"></a><span class="lineno"> 1375</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01376"></a><span class="lineno"> 1376</span> </div><div class="line"><a name="l01377"></a><span class="lineno"> 1377</span>  <span class="keywordflow">do</span> i = 1, len</div><div class="line"><a name="l01378"></a><span class="lineno"> 1378</span> <span class="comment">! Compute dda + ddb using Knuth's trick.</span></div><div class="line"><a name="l01379"></a><span class="lineno"> 1379</span>  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>  e = t1 - <span class="keywordtype">real</span>(dda(i))</div><div class="line"><a name="l01381"></a><span class="lineno"> 1381</span>  t2 = ((<span class="keywordtype">real(ddb(i))</span> - e) + (<span class="keywordtype">real(dda(i))</span> - (t1 - e))) &</div><div class="line"><a name="l01382"></a><span class="lineno"> 1382</span>  + aimag(dda(i)) + aimag(ddb(i))</div><div class="line"><a name="l01383"></a><span class="lineno"> 1383</span> </div><div class="line"><a name="l01384"></a><span class="lineno"> 1384</span> <span class="comment">! The result is t1 + t2, after normalization.</span></div><div class="line"><a name="l01385"></a><span class="lineno"> 1385</span>  ddb(i) = cmplx( t1 + t2, t2 - ((t1 + t2) - t1), r8 )</div><div class="line"><a name="l01386"></a><span class="lineno"> 1386</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01387"></a><span class="lineno"> 1387</span> </div><div class="line"><a name="l01388"></a><span class="lineno"> 1388</span> </div><div class="line"><a name="l01389"></a><span class="lineno"> 1389</span> <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> </div><div class="line"><a name="l01391"></a><span class="lineno"> 1391</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01392"></a><span class="lineno"> 1392</span> </div><div class="line"><a name="l01393"></a><span class="lineno"> 1393</span> <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> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01395"></a><span class="lineno"> 1395</span> <span class="comment">! </span></div><div class="line"><a name="l01396"></a><span class="lineno"> 1396</span> <span class="comment">! Purpose: </span></div><div class="line"><a name="l01397"></a><span class="lineno"> 1397</span> <span class="comment">! Split range into 'num_pieces'</span></div><div class="line"><a name="l01398"></a><span class="lineno"> 1398</span> <span class="comment">! </span></div><div class="line"><a name="l01399"></a><span class="lineno"> 1399</span> <span class="comment">!----------------------------------------------------------------------</span></div><div class="line"><a name="l01400"></a><span class="lineno"> 1400</span> </div><div class="line"><a name="l01401"></a><span class="lineno"> 1401</span> <span class="comment">! Arguments</span></div><div class="line"><a name="l01402"></a><span class="lineno"> 1402</span> </div><div class="line"><a name="l01403"></a><span class="lineno"> 1403</span>  <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>  <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>  <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> </div><div class="line"><a name="l01407"></a><span class="lineno"> 1407</span> <span class="comment">! Local workspace</span></div><div class="line"><a name="l01408"></a><span class="lineno"> 1408</span> </div><div class="line"><a name="l01409"></a><span class="lineno"> 1409</span>  <span class="keywordtype">integer</span> :: itmp1, itmp2, ioffset, i</div><div class="line"><a name="l01410"></a><span class="lineno"> 1410</span>  <span class="keywordtype">character(len=*)</span>,<span class="keywordtype">parameter</span> :: subname = <span class="stringliteral">'(oasis_reprosum_mod:split_indices)'</span></div><div class="line"><a name="l01411"></a><span class="lineno"> 1411</span> </div><div class="line"><a name="l01412"></a><span class="lineno"> 1412</span> <span class="comment">!-----------------------------------------------------------------------</span></div><div class="line"><a name="l01413"></a><span class="lineno"> 1413</span> </div><div class="line"><a name="l01414"></a><span class="lineno"> 1414</span>  itmp1 = total/num_pieces</div><div class="line"><a name="l01415"></a><span class="lineno"> 1415</span>  itmp2 = mod(total,num_pieces)</div><div class="line"><a name="l01416"></a><span class="lineno"> 1416</span>  ioffset = 0</div><div class="line"><a name="l01417"></a><span class="lineno"> 1417</span>  <span class="keywordflow">do</span> i=1,itmp2</div><div class="line"><a name="l01418"></a><span class="lineno"> 1418</span>  ibeg(i) = ioffset + 1</div><div class="line"><a name="l01419"></a><span class="lineno"> 1419</span>  iend(i) = ioffset + (itmp1+1)</div><div class="line"><a name="l01420"></a><span class="lineno"> 1420</span>  ioffset = iend(i)</div><div class="line"><a name="l01421"></a><span class="lineno"> 1421</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01422"></a><span class="lineno"> 1422</span>  <span class="keywordflow">do</span> i=itmp2+1,num_pieces</div><div class="line"><a name="l01423"></a><span class="lineno"> 1423</span>  ibeg(i) = ioffset + 1</div><div class="line"><a name="l01424"></a><span class="lineno"> 1424</span>  <span class="keywordflow">if</span> (ibeg(i) > total) <span class="keywordflow">then</span></div><div class="line"><a name="l01425"></a><span class="lineno"> 1425</span>  iend(i) = ibeg(i) - 1</div><div class="line"><a name="l01426"></a><span class="lineno"> 1426</span>  <span class="keywordflow">else</span></div><div class="line"><a name="l01427"></a><span class="lineno"> 1427</span>  iend(i) = ioffset + itmp1</div><div class="line"><a name="l01428"></a><span class="lineno"> 1428</span>  ioffset = iend(i)</div><div class="line"><a name="l01429"></a><span class="lineno"> 1429</span> <span class="keywordflow"> endif</span></div><div class="line"><a name="l01430"></a><span class="lineno"> 1430</span> <span class="keywordflow"> enddo</span></div><div class="line"><a name="l01431"></a><span class="lineno"> 1431</span> </div><div class="line"><a name="l01432"></a><span class="lineno"> 1432</span> <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> </div><div class="line"><a name="l01434"></a><span class="lineno"> 1434</span> <span class="comment">!========================================================================</span></div><div class="line"><a name="l01435"></a><span class="lineno"> 1435</span> </div><div class="line"><a name="l01436"></a><span class="lineno"> 1436</span> <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> |
---|