Re: [eigen] 3.0.4 coming soon, testing appreciated

[ Thread Index | Date Index | More Archives ]

On Mon, Nov 21, 2011 at 8:45 AM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> We would like to release Eigen 3.0.4 next week, so testing would be
> much appreciated on all supported platforms...

The following all look good on Ubuntu 11.04 x86_64:
   gcc 4.5.1
   icpc 10.1 20081024
   icpc 11.1 20100806

On a different system (an older Red Hat variant, x86_64), I see
   icpc 12.0.4 20110427
   icpc 12.1.0 20111011
both build all the 'make check' targets.  I noticed some terribly long
individual test times for those two builds, e.g.
   226/593 Test #226: product_trmm_32 ..................   Passed   19.90 sec
   228/593 Test #228: product_trmm_13 ..................   Passed  373.70 sec
   230/593 Test #230: product_trmm_23 ..................   Passed  453.71 sec,
which were not consistent with my gcc 4.5.1, intel 10.1, and intel
11.1 test runtimes.  Those long timings are suspect as I ran with
niceness +10 due to being on a login node on an HPC system.  I stopped
testing at this point.  Still, if anyone has access to a box with
those compiler versions, it might be worth checking those test
runtimes for either bad or regressed performance.

Finally, icpc 12.0.0 20101006 fails on Ubuntu 11.04 x86_64 with an
error message like...

[ 14%] Building CXX object test/CMakeFiles/exceptions.dir/exceptions.cpp.o
/workspace/rhys/eigen/Eigen/src/Core/Product.h(430): internal error:
assertion failed at: "shared/edgcpfe/exprutil.c", line 1490

      bool alphaIsCompatible = (!ComplexByReal) ? true :

compilation aborted for /workspace/rhys/eigen/test/exceptions.cpp (code 4)

I would be happy to try out any tweaks to get Intel 12.0.0 to pass
'make check'.  However, I suspect it's not worth the bother as it's an
"internal error" on Intel's part and it does not impact the latest
Intel 12.0.x release.  FWIW, I've Googled around for the error message
and the Intel articles that include 'internal error: assertion failed
at: "shared/edgcpfe/exprutil.c"' reference a problem that should sink
Intel 11.1 also.  Since Intel 11.1 built just fine I suspect some
other internal, version-specific Intel bug being at fault.

Hope this helps,

P.S. I know it's stupidly minor, but I can I pester someone to fix the
pkg-config-related regression ?  Release 3.0.3
"fixed" pkg-config for other folks but broke it for me.

Mail converted by MHonArc 2.6.19+