Recent changes to 556: test_matrix_multiply hangshttps://sourceforge.net/p/gnudatalanguage/bugs/556/2014-10-06T17:01:23Z#556 test_matrix_multiply hangs2014-10-06T17:01:23Z2014-10-06T17:01:23ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.netac8ebc24dccb8832fcc64f5a0256315c8774ac1f<div class="markdown_content"><p>I've been able to re-enable openmpi on Fedora 21+ and EPEL7 with eigen 3.2.2. So hopefully we're okay now.</p></div>#556 test_matrix_multiply hangs2014-06-25T07:09:37Z2014-06-25T07:09:37ZSylwester Arabashttps://sourceforge.net/u/slayoo/https://sourceforge.net9d6e4b947cd6bbaad05eb1a89038499abb3c4313<div class="markdown_content"><p>Hi,</p>
<p>For the record, I get the same behaviour on current Debian sid with cvs GDL and Eigen 3.2.1.</p>
<p>The test runs forever. Moreover, stopping (ctrl-c) the "make check" process leaves a 100% cpu gdl process in the background.</p>
<p>Compiling with OPENMP=NO makes it pass this test.</p>
<p>Sylwester</p></div>#556 test_matrix_multiply hangs2013-10-08T22:28:45Z2013-10-08T22:28:45ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.net9fdcc3f2b7d103d61c8b1ca51b22b7030106efda<div class="markdown_content"><p>Looks like things are happier if I disable openmp. Now to figure out why...</p></div>#556 test_matrix_multiply hangs2013-10-07T21:38:50Z2013-10-07T21:38:50Zgiloohttps://sourceforge.net/u/gilles-duvert/https://sourceforge.net2aa824b16539b72f08e338aa974a1d7e463d5149<div class="markdown_content"><p>indeed Dynamic and -1 are the same:</p>
<p>/usr/local/include/eigen3/Eigen/src/Core/util/Constants.h:<br />
(...)<br />
const int Dynamic = -1;</p></div>#556 test_matrix_multiply hangs2013-10-04T21:00:37Z2013-10-04T21:00:37ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.netf25ffeee6c62966b2686b160fe436d4c48668a74<div class="markdown_content"><p>Crap, or maybe not. Is -1 and Dynamic the same? Trouble is it seemed to fix it for my local builds, but not on the Fedora builders. I wonder though if this problem is then endemic to Eigen3</p></div>#556 test_matrix_multiply hangs2013-10-04T20:17:16Z2013-10-04T20:17:16ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.net37619ae293ca3fd082d6cf1c40d5ea6eaf8a0c84<div class="markdown_content"><p>Now that I've figured it out, it seems painfully obvious what the problem was and I'm upset that it took me so much time to do so. In my defence, I had never used the Eigen3 library before and had assumed it was being used correctly, but it was not. In basic_op.cpp dynamically sized matrices were declared:</p>
<div class="codehilite"><pre><span class="n">Matrix</span><span class="o"><</span><span class="n">Ty</span><span class="p">,</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span><span class="o">-</span><span class="mi">1</span><span class="o">></span>
</pre></div>
<p>but this is not correct. This declares a matrix the size of "-1" which can be interpreted by the compiler in very strange ways. The proper declaration is:</p>
<div class="codehilite"><pre><span class="n">Matrix</span><span class="o"><</span><span class="n">Ty</span><span class="p">,</span><span class="n">Dynamic</span><span class="p">,</span><span class="n">Dynamic</span><span class="o">></span>
</pre></div>
<p>The attached patch fixes. I'm sorry this didn't make it into 0.9.4, but I think this may warrant another quick release as the current code is seriously broken.</p></div>#556 test_matrix_multiply hangs2013-10-01T19:56:54Z2013-10-01T19:56:54ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.net2e365865553e2c64eda9373befd37a151d5dd4c7<div class="markdown_content"><p>Well, I've moved on to 0.9.4 - trying to package that up for Fedora. It's a really strange error that seems to come and go so it speaks to some kind of memory error.</p></div>#556 test_matrix_multiply hangs2013-10-01T19:30:10Z2013-10-01T19:30:10Zgiloohttps://sourceforge.net/u/gilles-duvert/https://sourceforge.netba54411fa7bcef043a2c762b9cffbd42606743d3<div class="markdown_content"><p>Hi,<br />
I am lost -- your test refers to 0.9.3, yes?<br />
The output says:<br />
"export/home/orion/fedora/gdl/<strong>gdl-0.9.3</strong>/build/testsuite/launchtest"</p>
<p>works for me with 0.9.4 and after...</p></div>#556 test_matrix_multiply hangs2013-10-01T18:20:02Z2013-10-01T18:20:02ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.nete1c34c336a38056e3c3aa4e78a8369cf0230af5d<div class="markdown_content"><p>To be more precise - gsl_linalg_LU_lndet() is returning -inf.</p></div>#556 test_matrix_multiply hangs2013-10-01T17:58:52Z2013-10-01T17:58:52ZOrion Poplawskihttps://sourceforge.net/u/opoplawski/https://sourceforge.net71442d463cb147279ac862ef955dfe583bd14c5e<div class="markdown_content"><p>Looks like invert is saying that the matrix is singular in these cases.</p></div>