[Q-e-developers] Using Valgrind and GDB with QE
Paolo Giannozzi
p.giannozzi at gmail.com
Thu May 5 19:44:09 CEST 2016
I have occasionally tried to run qe with valgrind. It gives a very large
number of "Invalid write of size 8" errors that do not seem to correspond
to anything significant. There are a few options, something like
--leak-check --show-origin, that yield a more readable summary.
Paolo
On Thu, May 5, 2016 at 7:31 PM, Ye Luo <xw111luoye at gmail.com> wrote:
> It doesn't seem to be a libc bug.
> The call stack shows that in the subroutine check_quantity_dh of your
> module mbdvdw,
> the code failed in writing something into tgg_complex which seems to be a
> variable belongs to your module.
> Is this variable dynamically allocatable and not initialized?
>
> Ye
>
> ===================
> Ye Luo, Ph.D.
> Leadership Computing Facility
> Argonne National Laboratory
>
> 2016-05-05 12:14 GMT-05:00 Hsin-Yu Ko <hsinyu at princeton.edu>:
>
>> Thomas,
>>
>> That is interesting. What you are seeing seems to be a libc bug [1]. I
>> have encountered something similar last month. I fixed the problem on my
>> machine by recompiling glibc with debug features enabled (I am not sure
>> how useful gentoo documentation is but I put the reference here just in
>> case [2]). I think removing -pg may be a quick fix according to [1].
>>
>> Best,
>> Hsin-Yu
>>
>> [1] http://valgrind-users.narkive.com/MPnV7HOw/gcc-pg-valgrind-errors
>> [2] https://wiki.gentoo.org/wiki/Debugging
>>
>> On 05/05/2016 12:34 PM, Thomas Markovich wrote:
>> > Hsin-Yu,
>> >
>> > Thank you for the suggestion!
>> >
>> > I had -g in LDFLAGS:
>> > LDFLAGS = -g -pthread -fopenmp
>> >
>> > but nothing equivalent in CFLAGS, which was defined as:
>> > CFLAGS = -O3 $(DFLAGS) $(IFLAGS)
>> >
>> > I have since gone ahead and changed CFLAGS to
>> > CFLAGS = -Og -g $(DFLAGS) $(IFLAGS)
>> >
>> > The resulting fortran compile statements look something like:
>> > mpif90 -Og -g -pg -fopenmp -Wall -Wextra -Warray-temporaries
>> > -Wconversion -fbacktrace -ffree-line-length-0 -finit-real=nan
>> > -ffpe-trap=zero,invalid,zero,overflow -x f95-cpp-input -fopenmp
>> > -D__GFORTRAN -D__STD_F95 -D__FFTW -D__MPI -D__PARA -D__SCALAPACK
>> > -D__OPENMP -I../include -I../iotk/src -I../ELPA/src -I. -c mbdvdw.f90
>> >
>> > and reran valgrind. It gave the following output:
>> >
>> > ==30486== Invalid write of size 8
>> > ==30486== at 0x1002735ED: __mbdvdw_module_MOD_mbdvdw_tgg_complex (in
>> > /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x100290F58:
>> > __mbdvdw_module_MOD_mbdvdw_check_quantity_dh (in
>> /Users/tmarkovich/bin/pw.x)
>> > ==30486== Address 0x1037989f0 is 0 bytes after a block of size 1,728
>> > alloc'd
>> > ==30486== at 0x10092B4AB: malloc (in
>> >
>> /usr/local/Cellar/valgrind/HEAD/lib/valgrind/vgpreload_memcheck-amd64-darwin.so)
>> > ==30486== by 0x10028FEE2:
>> > __mbdvdw_module_MOD_mbdvdw_check_quantity_dh (in
>> /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x1001D4567: v_of_rho_ (in /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x10007C0BE: electrons_scf_ (in
>> /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x10007D385: electrons_ (in /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x10018B30B: run_pwscf_ (in /Users/tmarkovich/bin/pw.x)
>> > ==30486== by 0x100001157: MAIN__ (pwscf.f90:30)
>> > ==30486== by 0x1004EC496: main (pwscf.f90:14)
>> >
>> > This appears to not have changed much.
>> >
>> > Best,
>> > Thomas
>> >
>> > On Thu, May 5, 2016 at 10:15 AM, Hsin-Yu Ko <hsinyu at princeton.edu
>> > <mailto:hsinyu at princeton.edu>> wrote:
>> >
>> > Hi Thomas,
>> >
>> > Did you put -g in CFLAGS and LDFLAGS? Valgrind seems to recognize
>> some
>> > lines inside MAIN__ while failing to find the linked ones.
>> >
>> > Best,
>> > Hsin-Yu
>> >
>> > On 05/05/2016 09:48 AM, Thomas Markovich wrote:
>> > > Hi,
>> > >
>> > > I'm preparing to push my module that implements the Many Body
>> > Dispersion
>> > > van der Waals correction, and all associated forces. As a last
>> > thing, I
>> > > ran my code through valgrind, and it seems to have popped up a
>> > couple of
>> > > remaining things that I would like to fix before release[1].
>> > > Unfortunately, the valgrind output below is less than clear on
>> where
>> > > exactly the error is, and it doesn't give any important line
>> numbers.
>> > > Beyond this, addr2line gives thoroughly unhelpful output:
>> > > ▶ gaddr2line -e pw.x 0x100520518
>> > > ??:0.
>> > >
>> > > I have compiled QE given the following flags with gfortran 4.9:
>> > > FFLAGS = -Og -g -pg -fopenmp -fbacktrace -fcheck=all
>> > > -finit-real=nan -ffpe-trap=zero,invalid,zero,overflow
>> > >
>> > > Is there any way to compile QE such that it generates all the
>> > debugging
>> > > symbols, so that I can get more readable and informative output
>> from
>> > > valgrind? I thought all I needed was the -g flag, but it appears
>> > that I
>> > > might need more?
>> > >
>> > > Best,
>> > > Thomas Markovich
>> > >
>> > > [1]
>> > > ==10233== Invalid write of size 8
>> > > ==10233== at 0x100520518:
>> > __mbdvdw_module_MOD_mbdvdw_tgg_complex (in
>> > > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x100555A88:
>> > > __mbdvdw_module_MOD_mbdvdw_check_quantity_dh (in
>> > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== Address 0x103131248 is 8 bytes after a block of size
>> 1,728
>> > > alloc'd
>> > > ==10233== at 0x1011814AB: malloc (in
>> > >
>> >
>> /usr/local/Cellar/valgrind/HEAD/lib/valgrind/vgpreload_memcheck-amd64-darwin.so)
>> > > ==10233== by 0x1005547B0:
>> > > __mbdvdw_module_MOD_mbdvdw_check_quantity_dh (in
>> > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x1003D56E4: v_of_rho_ (in
>> /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x1000F540B: electrons_scf_ (in
>> > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x1000F6E18: electrons_ (in
>> > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x10032AA28: run_pwscf_ (in
>> > /Users/tmarkovich/bin/pw.x)
>> > > ==10233== by 0x1000010BB: MAIN__ (pwscf.f90:30)
>> > > ==10233== by 0x100B67C1F: main (pwscf.f90:14)
>> > _______________________________________________
>> > Q-e-developers mailing list
>> > Q-e-developers at qe-forge.org <mailto:Q-e-developers at qe-forge.org>
>> > http://qe-forge.org/mailman/listinfo/q-e-developers
>> >
>> >
>> _______________________________________________
>> Q-e-developers mailing list
>> Q-e-developers at qe-forge.org
>> http://qe-forge.org/mailman/listinfo/q-e-developers
>>
>
>
> _______________________________________________
> Q-e-developers mailing list
> Q-e-developers at qe-forge.org
> http://qe-forge.org/mailman/listinfo/q-e-developers
>
>
--
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20160505/9cb3ea87/attachment.html>
More information about the developers
mailing list