[Llvm-bgq-discuss] bgclang r185415-20130701 on vesta/mira

Erik Schnetter schnetter at cct.lsu.edu
Fri Jul 5 14:15:18 CDT 2013


Hal

The failing code is now in ~eschnett/vecmathlib on Vesta. The following
sequence of commands reproduces this problem (lengthy on purpose):

/home/projects/llvm/bin/bgclang++11
-I/soft/compilers/ibmcmp-may2013/xlmass/bg/7.3/include -E test.cc > test.i
/home/projects/llvm/bin/bgclang++11 -S test.i
/home/projects/llvm/bin/bgclang++11 -c test.s
/home/projects/llvm/bin/bgclang++11 -o test test.o
-L/soft/compilers/ibmcmp-may2013/xlmass/bg/7.3/bglib64 -lmassv -lmass
-lmass_simd
qsub -A DynamicDevel -n 1 -t 10 ./test

That is, you can either look at test.cc, or at the preprocessed test.i to
see the problem. The current output is located in the file "106581.output".

The screen output will be somewhat jumbled -- sorry, I can clean this up.
The screen output indicates that a vector4double containing all false (all
elements -1.0) has operator[] applied to all elements, which works
correctly. These are then converted to bool (by a routine of mine), which
works fine -- these are the lines showing "[]=0" where 0 indicates false.
The line "fvml(x)=1 [0x01]" then shows that the result of the 3 scalar &&
operations yields true.

If the error is not present, then the line "Error in all(false):" should
disappear.

The relevant code that fails is the routine all() in the file
vec_qpx_double4.h. This routine calls operator[] on a vector, and then
reduces the elements via the && operators.

Let me know if you need more information.

-erik



On Fri, Jul 5, 2013 at 2:57 PM, Hal Finkel <hfinkel at anl.gov> wrote:

> ----- Original Message -----
> >
> > Hal
> >
> >
> > I this version supposed to handle the && operator correctly?
>
> Yes, I fixed the bug responsible for that issue.
>
> >
> >
> > I believe that my original test case now works fine, but I encounter
> > a different problem with the && and || operators, even without
> > optimization.
>
> Can you point me to the failing code and I'll take a look.
>
> Thanks again,
> Hal
>
> >
> >
> > -erik
> >
> >
> >
> >
> >
> > On Tue, Jul 2, 2013 at 12:07 PM, Hal Finkel < hfinkel at anl.gov >
> > wrote:
> >
> >
> > Hello everyone,
> >
> > I've upgraded the bgclang install on vesta/mira to build
> > r185415-20130701. This build incorporates a number of important bug
> > fixes, and while hitting these bugs seems rare, I recommend that all
> > codes compiled with previous builds be recompiled.
> >
> > For those of you maintaining your own installs, I've posted the
> > updated patches and build/wrapper scripts on:
> > https://trac.alcf.anl.gov/projects/llvm-bgq
> >
> > -Hal
> >
> > --
> > Hal Finkel
> > Assistant Computational Scientist
> > Leadership Computing Facility
> > Argonne National Laboratory
> > _______________________________________________
> > llvm-bgq-discuss mailing list
> > llvm-bgq-discuss at lists.alcf.anl.gov
> > https://lists.alcf.anl.gov/mailman/listinfo/llvm-bgq-discuss
> >
> >
> >
> >
> > --
> > Erik Schnetter < schnetter at cct.lsu.edu >
> > http://www.perimeterinstitute.ca/personal/eschnetter/
>
> --
> Hal Finkel
> Assistant Computational Scientist
> Leadership Computing Facility
> Argonne National Laboratory
>



-- 
Erik Schnetter <schnetter at cct.lsu.edu>
http://www.perimeterinstitute.ca/personal/eschnetter/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alcf.anl.gov/pipermail/llvm-bgq-discuss/attachments/20130705/64ed3e26/attachment.html>


More information about the llvm-bgq-discuss mailing list