<html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body bgcolor="#FFFFFF" text="#000000">
I'm currently working on some performance numbers of MPICH 3.1.3 on
the Tilera TILE-Gx platform. While direct support isn't indicated, I
wanted to see what the performance profile would be for some of the
core MPI functions. The barrier times, however, exhibit poor
performance and high variance as the number of PEs scale up on this
36-core SMP device.<br>
<div class="moz-forward-container"> <br>
<blockquote><tt>PEs Trial 1 Trial 2</tt><tt> (times in
microseconds)<br>
</tt><tt>2 7.28 7.54</tt><tt><br>
</tt><tt>3 11.63 13010.92</tt><tt><br>
</tt><tt>4 15.01 268</tt><tt><br>
</tt><tt>5 21073.14 18.11</tt><tt><br>
</tt><tt>6 19.63 18.97</tt><tt><br>
</tt><tt>7 17.83 22698.7</tt><tt><br>
</tt><tt>8 23.82 4099.2</tt><tt><br>
</tt><tt>9 24.72 9334.79</tt><tt><br>
</tt><tt>10 39859.54 1591.71</tt><tt><br>
</tt><tt>11 39586.55 38599.86</tt><tt><br>
</tt><tt>12 30229.56 39799.5</tt><tt><br>
</tt><tt>13 28860.67 5684.71</tt><tt><br>
</tt><tt>14 38742.38 39956.55</tt><tt><br>
</tt><tt>15 43117.49 39952.83</tt><tt><br>
</tt><tt>16 26762.15 26230.78</tt><tt><br>
</tt><tt>17 39458.78 59911.38</tt><tt><br>
</tt><tt>18 46232.86 45327.22</tt><tt><br>
</tt><tt>19 45462.67 54762.76</tt><tt><br>
</tt><tt>20 59744.57 39657.49</tt><tt><br>
</tt><tt>21 54023.2 72209.07</tt><tt><br>
</tt><tt>22 78213.27 49108.57</tt><tt><br>
</tt><tt>23 69221.4 79534.28</tt><tt><br>
</tt><tt>24 71599.46 81920.28</tt><tt><br>
</tt><tt>25 45443.61 76059.46</tt><tt><br>
</tt><tt>26 50649.56 72402.17</tt><tt><br>
</tt><tt>27 39936.37 66906.83</tt><tt><br>
</tt><tt>28 64324.64 44803.16</tt><tt><br>
</tt><tt>29 59913.22 53071.9</tt><tt><br>
</tt><tt>30 44569.65 70844.15</tt><tt><br>
</tt><tt>31 55428.5 47399.51</tt><tt><br>
</tt><tt>32 54708.82 64102.45</tt><tt><br>
</tt><tt>33 34413.99 52384.35</tt><tt><br>
</tt><tt>34 66390.67 97134.79</tt><tt><br>
</tt><tt>35 68839.46 62719.52</tt><tt><br>
</tt><tt>36 43243.98 59132.83</tt><tt> (59.1 ms)<br>
</tt></blockquote>
Each entry is an average of 100 iterations. Benchmark used is from
OSU MPI microbenchmarks (osu_barrier.c).<br>
<br>
These are the configure options I used to cross-compile and build
MPICH for TILE-Gx:<br>
<blockquote><tt>mkdir build && cd build</tt><br>
<tt>PREFIXROOT=/opt/tilegx</tt><br>
<tt>../configure CC=tile-gcc CXX=tile-g++ FC=tile-gfortran
F77=tile-gfortran \</tt><br>
<tt> --prefix=$PREFIXROOT/mpich --host=tile \</tt><br>
<tt> --disable-fortran \</tt><br>
<tt> --enable-fast=yes --with-thread-package=posix</tt><br>
</blockquote>
<br>
Are there any build options that I'm not using that I should be?
I'm trying to focus this build specifically for SMP performance.<br>
<br>
Thanks,<br>
<br>
Bryant Lam<br>
<br>
<br>
<br>
<br>
</div>
<br>
</body>
</html>