sorry for not coming back on that thread before.<br>I do agree on the lack of documentation, especially on sparse matrices.<br><br>But... tada... as I am working on improving this documentation now, I would appreciate if you guys can send me examples of nice tricks when using sparse matrices of all sorts. Any examples are welcome and I will compile them into the new documentation.<br>
<br>If you have web pages and/or code that you can communicate, it will be greatly appreciated too.<br>I must say that I use dense matrices most of the time, so contributions are most welcome on sparse matrices.<br><br>In order not to bother the mailing list, you can also send all the materials to my personal email address directly.<br>
Thanks everybody.<br><br>Cheers,<br>David<br><br><div class="gmail_quote">On Thu, Apr 22, 2010 at 02:10, Sunil Thomas <span dir="ltr">&lt;<a href="mailto:sgthomas27@gmail.com">sgthomas27@gmail.com</a>&gt;</span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><div>Hi Nasos,</div>
<div> </div>
<div>I see...thanks a lot for testing this out..really appreciate! </div>
<div>I am using Boost 1.36.0 here (not 1.34 as I said before - sorry). </div>
<div>And yes, I do have -DNDEBUG option on all the time.</div>
<div> </div>
<div>Cheers,</div>
<div>Sunil.<font color="#888888"><br><br></font></div><div><div></div><div class="h5">
<div class="gmail_quote">On Wed, Apr 21, 2010 at 4:11 PM, Nasos Iliopoulos <span dir="ltr">&lt;<a href="mailto:nasos_i@hotmail.com" target="_blank">nasos_i@hotmail.com</a>&gt;</span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0px 0px 0px 0.8ex; padding-left: 1ex;">
<div>Thomas,<br><br>I tested your code and it looks that it runs for both compressed and mapped matrices (using -DNDEBUG, Boost v 1.40). ALTHOUGH  for mapped matrices without -DNDEBUG it throws from an assertion in the iterator. It looks like a bug in uBlas (maybe wrong assertion), but I am not definite about it. I will file a bug report if I find out. The code I used to test is:<br>

<br>
<p style="margin: 0px; text-indent: 0px;">#define NDEBUG</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">#include &lt;boost/numeric/ublas/matrix_sparse.hpp&gt;</p>
<p style="margin: 0px; text-indent: 0px;">#include &lt;boost/numeric/ublas/io.hpp&gt;</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">//#define USE_MAPPED</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">using namespace boost::numeric::ublas;</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">#ifdef USE_MAPPED</p>
<p style="margin: 0px; text-indent: 0px;">typedef mapped_matrix&lt;double&gt; matrix_type;</p>
<p style="margin: 0px; text-indent: 0px;">#else</p>
<p style="margin: 0px; text-indent: 0px;">typedef compressed_matrix&lt;double&gt; matrix_type;</p>
<p style="margin: 0px; text-indent: 0px;">#endif</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">int main(int argc, char *argv[])</p>
<p style="margin: 0px; text-indent: 0px;">{</p>
<p style="margin: 0px; text-indent: 0px;">const std::size_t size = 10;</p>
<p style="margin: 0px; text-indent: 0px;">matrix_type P(size,size);</p>
<p style="margin: 0px; text-indent: 0px;">for (int ii=0; ii!=P.size1(); ii++)</p>
<p style="margin: 0px; text-indent: 0px;">for (int j=0; j!=P.size2(); j++)</p>
<p style="margin: 0px; text-indent: 0px;">P(ii,j)=ii*j+ii+j;</p>
<p style="margin: 0px; text-indent: 0px;">std::cout &lt;&lt; P &lt;&lt; std::endl;</p>
<p style="margin: 0px; text-indent: 0px;">typedef matrix_type::reverse_iterator1 itmr1;</p>
<p style="margin: 0px; text-indent: 0px;">typedef matrix_type::iterator2 itm2;</p>
<p style="margin: 0px; text-indent: 0px;">int i,j ;</p>
<div>
<p style="margin: 0px; text-indent: 0px;">for(itmr1 i1 = P.rbegin1(); i1 != P.rend1(); ++i1) {</p>
<p style="margin: 0px; text-indent: 0px;">itm2 i2 = i1.begin(); i = (int)i2.index1();</p>
<p style="margin: 0px; text-indent: 0px;">for(; i2 != i1.end(); ++i2) {</p>
<p style="margin: 0px; text-indent: 0px;">j = (int) i2.index2();</p>
<p style="margin: 0px; text-indent: 0px;">if(j &lt;= i) continue;</p>
<p style="margin: 0px; text-indent: 0px;">//x(i) -= (*i2)*x(j);</p>
<p style="margin: 0px; text-indent: 0px;">}</p>
<p style="margin: 0px; text-indent: 0px;"></p></div>
<p style="margin: 0px; text-indent: 0px;">std::cout &lt;&lt; P(i,i) &lt;&lt; &quot; &quot;;</p>
<p style="margin: 0px; text-indent: 0px;">//x(i) *= P(i,i);</p>
<p style="margin: 0px; text-indent: 0px;"></p>
<p style="margin: 0px; text-indent: 0px;">}</p>
<p style="margin: 0px; text-indent: 0px;">return 0;</p>
<p style="margin: 0px; text-indent: 0px;">}</p>
<p style="margin: 0px; text-indent: 0px;"></p><br>Best<br>Nasos<br><br>
<hr>
Date: Wed, 21 Apr 2010 15:11:48 -0700 
<div>
<div></div>
<div><br>From: <a href="mailto:sgthomas27@gmail.com" target="_blank">sgthomas27@gmail.com</a><br>To: <a href="mailto:ublas@lists.boost.org" target="_blank">ublas@lists.boost.org</a><br>Subject: Re: [ublas] Question on ublas performance (resending as subscriber).<br>

<br>
<div>Thanks Nasos - I was going to try that next anyhow. Will let you know if I get a major improvement. </div>
<div> </div>
<div>In a related problem, I find that reverse iterator is not working as expected.. (for compressed_matrix </div>
<div>and mapped_matrix). I really doubt that it is a problem of my usage (but that is possible). Here is a </div>
<div>snippet of my usage from ILU back substitution (P is the matrix - compressed, or mapped):</div>
<div> </div><font size="2" color="#008000"><font size="2" color="#008000">for(itmr1 i1 = P.rbegin1(); i1 != P.rend1(); ++i1) { <br>    itm2 i2 = i1.begin(); i = (int)i2.index1(); <br>    for(; i2 != i1.end(); ++i2) { <br>

          j = (int) i2.index2(); <br>          if(j &lt;= i) continue; <br>          x(i) -= (*i2)*x(j); <br>    }<br>    x(i) *= P(i,i);<br>}<br>
<div></div></font></font>Greatly appreciate if anybody knows why this crashes as soon as it enters outer loop...let me know if you need 
<div>any further info. Its been a while since I used it, I had given up on it and used a combination of a forward iterator </div>
<div>loop and a usual nested loop inside it. I have no such problem with forward iterator.</div>
<div> </div>
<div>Thanks in advance,</div>
<div>Sunil.</div>
<div> </div>
<div><br> </div>
<div>On Wed, Apr 21, 2010 at 2:53 PM, Nasos Iliopoulos <span dir="ltr">&lt;<a href="mailto:nasos_i@hotmail.com" target="_blank">nasos_i@hotmail.com</a>&gt;</span> wrote:<br>
<blockquote style="border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
<div>Thomas,<br><br>replace sparse_matrix with compressed_matrix. Also try the following:<br><br>use the generalized vector of compressed_vector A as in the link to build the stiffness matrix, I think this is faster than every other sparse container (mapped, compressed, coordinate). If you need a storage type like compressed (most likely if you use a classical sparse solver), just assign it after the generalized is filled by your stiffness matrix construction algorithm like:<br>

compressed_matrix&lt;double&gt; K=A;<br><br>(I am unsure that in earlier versions of uBlas the above is optimum, the latest svn though should be ok. What version are you using?)<br> <br>The downside of this approach is that it requires twice the memory.<br>

<br>Let us know if it works for you.<br><br>Best<br>Nasos<br><br>
<hr>
Date: Wed, 21 Apr 2010 13:51:56 -0700<br>From: <a href="mailto:sgthomas27@gmail.com" target="_blank">sgthomas27@gmail.com</a><br>To: <a href="mailto:ublas@lists.boost.org" target="_blank">ublas@lists.boost.org</a><br>Subject: Re: [ublas] Question on ublas performance (resending as subscriber). 
<div>
<div></div>
<div><br><br>
<div>Hi Jorn,</div>
<div> </div>
<div>Your answer correctly identified the problem. It appears for my problem, mapped_matrix is a </div>
<div>much better (by factor of 110), if not the best choice as far as assembling goes. It also hasn&#39;t </div>
<div>hurt traversal that badly, it appears</div>
<div> </div>
<div>In the link you attached (thank you!), the author uses something called &quot;sparse_matrix&quot; in </div>
<div>boost::numeric::ublas - does this even exist? Atleast in version 1.34, it gives me a compile </div>
<div>error, saying no sparse_matrix type exists in boost::numeric::ublas (it was the first thing I </div>
<div>tested before going for mapped_matrix) and the documentation online doesn&#39;t mention </div>
<div>sparse_matrix as a type of sparse storage at all. </div>
<div> </div>
<div>Anyway, thanks a lot for all of your help!</div>
<div> </div>
<div>Regards,</div>
<div>Sunil.<br><br></div>
<div>2010/4/21 Jörn Ungermann <span dir="ltr">&lt;<a href="mailto:j.ungermann@fz-juelich.de" target="_blank">j.ungermann@fz-juelich.de</a>&gt;</span><br>
<blockquote style="border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">Hi Sunil,<br><br>this is likely not a problem of uBLAS, but one of the principal problems<br>of using sparse matrices. Depending on the type of matrix either random<br>

access or multiplication performance is efficient.<br>For the compressed_matrix, random access is rather costly *unless* you<br>can control the way in which elements are added to the matrix. If you<br>can assemble the (row_major) compressed_matrix row-by-row with ascending<br>

column indices, this should take no time at all.<br>If you can&#39;t do this, use a different matrix type for assembly, e.g.<br>mapped_matrix (which offers efficient random access, but bad<br>computational performance) and construct the compressed_matrix from<br>

there.<br><br>See Gunter Winkler&#39;s page for details:<br><a href="http://www.guwi17.de/ublas/matrix_sparse_usage.html" target="_blank">http://www.guwi17.de/ublas/matrix_sparse_usage.html</a><br><br>Kind reagrds,<br>Jörn<br>


<div>
<div></div>
<div><br>On Wed, 2010-04-21 at 03:09 +0200, Sunil Thomas wrote:<br>&gt; Hi all,<br>&gt;<br>&gt;    I&#39;ve been using boost 1.34 ublas library, especially the<br>&gt; compressed_matrix class for sparse matrices in<br>&gt;<br>

&gt; compressed row storage form. But I noticed that simply accessing an<br>&gt; element of the matrix (to assign<br>&gt;<br>&gt; it a value, for example) slows my application down to unusable levels,<br>&gt; for problems of the order of just<br>

&gt;<br>&gt; 80,000 unknowns. I&#39;ve identified the program is there and yes, I am<br>&gt; allocating the memory as I should<br>&gt;<br>&gt; be for the matrix, - for example here is a snippet (of important<br>&gt; lines):<br>

&gt;<br>&gt;<br>&gt; ************************************************************************************<br>&gt; matrix_A = compressed_matrix(nelem_a(), nelem_a(), nonzeros());    //<br>&gt; allocation<br>&gt;<br>&gt; matrix_A(uic1, uic1) += -trans; // assignment<br>

&gt;<br>&gt; matrix_A(uic2, uic2) += -trans;  // assignment<br>&gt;<br>&gt; ************************************************************************************<br>&gt;<br>&gt; where all variables (and/or functions), e.g. uic1, uic2, trans,<br>

&gt; neleme_a(), nonzeros(), etc.. are all well-defined<br>&gt;<br>&gt; (this is all been checked thoroughly). Commenting  out the two<br>&gt; assignment statements for example reduced<br>&gt;<br>&gt; my overall run time from 110 seconds to 0 (practically zero), for<br>

&gt; 80000 runs. Has anyone encountered this<br>&gt;<br>&gt; problem and know of a solution? I&#39;ve heard a lot of stories about how<br>&gt; boost::ublas is just not up there in<br>&gt;<br>&gt; performance and I certainly hope I am missing something trivial. Do<br>

&gt; later versions of boost address this<br>&gt;<br>&gt; better?<br>&gt;<br>&gt;<br>&gt;<br>&gt; Greatly appreciate any help.<br>&gt;<br>&gt; Thanks,<br>&gt;<br>&gt; Sunil.<br>&gt;<br>&gt;<br><br><br></div></div>------------------------------------------------------------------------------------------------<br>

------------------------------------------------------------------------------------------------<br>Forschungszentrum Juelich GmbH<br>52425 Juelich<br>Sitz der Gesellschaft: Juelich<br>Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498<br>

Vorsitzende des Aufsichtsrats: MinDir&#39;in Baerbel Brumme-Bothe<br>Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender),<br>Dr. Ulrich Krafft (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,<br>Prof. Dr. Sebastian M. Schmidt<br>

------------------------------------------------------------------------------------------------<br>------------------------------------------------------------------------------------------------<br>_______________________________________________<br>

ublas mailing list<br><a href="mailto:ublas@lists.boost.org" target="_blank">ublas@lists.boost.org</a><br><a href="http://lists.boost.org/mailman/listinfo.cgi/ublas" target="_blank">http://lists.boost.org/mailman/listinfo.cgi/ublas</a><br>

Sent to: <a href="mailto:sgthomas27@gmail.com" target="_blank">sgthomas27@gmail.com</a></blockquote></div><br><br></div></div>
<div>
<hr>
The New Busy is not the too busy. Combine all your e-mail accounts with Hotmail. <a href="http://www.windowslive.com/campaign/thenewbusy?tile=multiaccount&amp;ocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_4" target="_blank">Get busy.</a></div>

</div><br>_______________________________________________<br>ublas mailing list<br><a href="mailto:ublas@lists.boost.org" target="_blank">ublas@lists.boost.org</a><br><a href="http://lists.boost.org/mailman/listinfo.cgi/ublas" target="_blank">http://lists.boost.org/mailman/listinfo.cgi/ublas</a><br>

Sent to: <a href="mailto:sgthomas27@gmail.com" target="_blank">sgthomas27@gmail.com</a><br></blockquote></div><br><br>
<hr>
The New Busy is not the too busy. Combine all your e-mail accounts with Hotmail. <a href="http://www.windowslive.com/campaign/thenewbusy?tile=multiaccount&amp;ocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_4" target="_blank">Get busy.</a></div>

</div></div><br>_______________________________________________<br>ublas mailing list<br><a href="mailto:ublas@lists.boost.org" target="_blank">ublas@lists.boost.org</a><br><a href="http://lists.boost.org/mailman/listinfo.cgi/ublas" target="_blank">http://lists.boost.org/mailman/listinfo.cgi/ublas</a><br>

Sent to: <a href="mailto:sgthomas27@gmail.com" target="_blank">sgthomas27@gmail.com</a><br></blockquote></div><br>
</div></div><br>_______________________________________________<br>
ublas mailing list<br>
<a href="mailto:ublas@lists.boost.org">ublas@lists.boost.org</a><br>
<a href="http://lists.boost.org/mailman/listinfo.cgi/ublas" target="_blank">http://lists.boost.org/mailman/listinfo.cgi/ublas</a><br>
Sent to: <a href="mailto:david.bellot@gmail.com">david.bellot@gmail.com</a><br></blockquote></div><br><br clear="all"><br>-- <br>David Bellot, PhD<br><a href="mailto:david.bellot@gmail.com">david.bellot@gmail.com</a><br>
<a href="http://david.bellot.free.fr">http://david.bellot.free.fr</a><br><br>