Re: [eigen] Re: pre-RC1 testing needed

[Cyrille Berger <cberger@xxxxxxxxxxx>]

On Monday 26 January 2009, Benoit Jacob wrote:
> Keir: indeed good idea, it's always useful to hear positive reports too.
>
> Oh, and in case what's blocking people is that you don't know how to
> build and run the tests, here you go:
>
> 1) checkout the latest sources
>
> svn co svn://anonsvn.kde.org/home/kde/trunk/kdesupport/eigen2
>
> 2) create a build directory "eigen2build" alongside the "eigen2" directory
>
> mkdir eigen2build
>
> 3) enter the build dir
>
> cd eigen2build
>
> 4) run CMake, enable SSE2 or AltiVec testing (only used if available
> so you can always enable that)
>
> cmake -DEIGEN_BUILD_TESTS=ON -DEIGEN_TEST_SSE2=ON ../eigen2
>
> 5) run make, with an appropriate number of jobs: each job can take up
> to 450 MB of RAM with GCC and up to 800 MB of RAM with MSVC
>
> make -j2 // for 2 jobs
>
> 6) run the tests
>
> cd test
> ctest
>
> 7) if you want, you can run a test individually. You can get command
> line help too.
>
> ./test_lu r1 s12345678
>
> Cheers,
> Benoit
>
> 2009/1/26 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
> > Hi,
> >
> > As you could see, there were a lot more changes than expected going
> > into RC1. Now we think it's ready, but we aren't able to test on many
> > platforms and typically it's expected that the recent changes may
> > cause minor breakage (like compilation errors) on some platforms.
> >
> > So testing is much appreciated.
> > - build and run unit tests
> > - build and run your own software using eigen
> >
> > We are specifically interested in the following setups:
> > * MSVC 2008 with SSE2 enabled
> > * MSVC 2005 (SSE isn't used here)
> > * Old GCC (from 3.3 to 4.1) (SSE isn't used here) including, but not
> > only, MinGW * ICC, recent versions, with SSE2
> > * Power machine with AltiVec enabled
> > * BSD (Free/Open/Net/PC) with SSE2 (if you have GCC >= 4.2) or AltiVec
> > enabled
> >
> > No unit test is supposed to fail!!
> >
> > Cheers,
> > Benoit
On FreeBSD x86-64 (apart of ctest not willing to work...):

./test_dynalloc
Initializing random number generator with seed 1233528847
Repeating each test 10 times
Test check_aligned_malloc() failed in "/home/cyrille/tmp/eigen2/test/dynalloc.cpp" (44)
size_t(p)%16==0

I am not sure what made you think FreeBSD was returning alligned memory with malloc ? But the attached patch fixed dynalloc for me.
For NetBSD I will try to think to ask one of my coworker who happen to be a NetBSD developer.

--
Cyrille Berger

Index: Eigen/src/Core/util/Memory.h
===================================================================
--- Eigen/src/Core/util/Memory.h	(revision 919950)
+++ Eigen/src/Core/util/Memory.h	(working copy)
@@ -28,7 +28,7 @@
 #define EIGEN_MEMORY_H
 
 // for NetBSD I didn't see any clear statement in the docs, but Mark Davies is confident about this.
-#if defined(__APPLE__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(_WIN64)
+#if defined(__APPLE__) || defined(__NetBSD__) || defined(_WIN64)
   #define EIGEN_MALLOC_ALREADY_ALIGNED 1
 #else
   #define EIGEN_MALLOC_ALREADY_ALIGNED 0