[petsc-dev] [petsc-maint] Incident INC0122538 MKL on Cori/KNL
Smith, Barry F.
bsmith at mcs.anl.gov
Tue Jul 3 14:28:55 CDT 2018
Could also be a bug in the new hash table stuff for options?
> ==3965== valgrind: Unrecognised instruction at address 0x20104a53.
> ==3965== at 0x20104A53: kh_resize_HTPrinted (viewreg.c:12)
> ==3965== by 0x20104EE2: kh_put_HTPrinted (viewreg.c:12)
==3965== by 0x201061EF: PetscOptionsHelpPrintedCheck (viewreg.c:89)
Barry
> On Jul 3, 2018, at 1:10 PM, Mills, Richard Tran <rtmills at anl.gov> wrote:
>
> Mark, were you trying this in Valgrind with a binary targeting KNL, i.e., built to use AVX-512 instructions? I don't think Valgrind implements all (or any?) of those, so a failure is not a surprise. Indeed, I've had Valgrind choke on some AVX2 instructions, though maybe the most recent versions of Valgrind will handle these properly now.
>
> --Richard
>
> On Tue, Jul 3, 2018 at 4:25 AM, Mark Adams <mfadams at lbl.gov> wrote:
> Well this does work without valgrind.
>
> On Tue, Jul 3, 2018 at 6:36 AM Mark Adams <mfadams at lbl.gov> wrote:
> I built a 32 bit integer version and now it dies in PetscInit. Ugh.
>
> ==3965== Conditional jump or move depends on uninitialised value(s)
> ==3965== at 0x27AFFD8F: _int_free (malloc.c:3945)
> ==3965== by 0x20074F48: PetscOptionsSetValue (options.c:1152)
> ==3965== by 0x20070450: PetscOptionsInsertArgs_Private (options.c:636)
> ==3965== by 0x2007189F: PetscOptionsInsert (options.c:746)
> ==3965== by 0x20093DB1: PetscInitialize (pinit.c:929)
> ==3965== by 0x2000ABCE: main (ex19.c:106)
> ==3965==
> vex amd64->IR: unhandled instruction bytes: 0x62 0xF1 0x7F 0x8 0x7B 0xC2 0xC5 0xFB 0x10 0xD
> vex amd64->IR: REX=0 REX.W=0 REX.R=0 REX.X=0 REX.B=0
> vex amd64->IR: VEX=0 VEX.L=0 VEX.nVVVV=0x0 ESC=NONE
> vex amd64->IR: PFX.66=0 PFX.F2=0 PFX.F3=0
> ==3965== valgrind: Unrecognised instruction at address 0x20104a53.
> ==3965== at 0x20104A53: kh_resize_HTPrinted (viewreg.c:12)
> ==3965== by 0x20104EE2: kh_put_HTPrinted (viewreg.c:12)
> ==3965== by 0x201061EF: PetscOptionsHelpPrintedCheck (viewreg.c:89)
> ==3965== by 0x2009AD61: PetscOptionsBegin_Private (aoptions.c:34)
> ==3965== by 0x2008362E: PetscOptionsSetFromOptions (options.c:2646)
> ==3965== by 0x20094689: PetscInitialize (pinit.c:967)
> ==3965== by 0x2000ABCE: main (ex19.c:106)
> ==3965== Your program just tried to execute an instruction that Valgrind
> ==3965== did not recognise. There are two possible reasons for this.
> ==3965== 1. Your program has a bug and erroneously jumped to a non-code
> ==3965== location. If you are running Memcheck and you just saw a
> ==3965== warning about a bad jump, it's probably your program's fault.
> ==3965== 2. The instruction is legitimate but Valgrind doesn't handle it,
> ==3965== i.e. it's Valgrind's fault. If you think this is the case or
> ==3965== you are not sure, please let us know and we'll try to fix it.
> ==3965== Either way, Valgrind will now raise a SIGILL signal which will
> ==3965== probably kill your program.
> [0]PETSC ERROR: ==3965== Conditional jump or move depends on uninitialised value(s)
> ==3965== at 0x27B18B48: strchrnul (strchr.S:106)
> ==3965== by 0x27AE30C8: __find_specmb (printf-parse.h:108)
> ==3965== by 0x27AE30C8: vfprintf (vfprintf.c:1311)
> ==3965== by 0x27AFA045: vsnprintf (vsnprintf.c:119)
> ==3965== by 0x20116802: PetscVSNPrintf (mprint.c:178)
> ==3965== by 0x20117006: PetscVFPrintfDefault (mprint.c:294)
> ==3965== by 0x20136BB2: PetscErrorPrintfDefault (errtrace.c:116)
> ==3965== by 0x2013938C: PetscSignalHandlerDefault (signal.c:131)
> ==3965== by 0x2013903B: PetscSignalHandler_Private (signal.c:43)
> ==3965== by 0x2335D07F: ??? (in /global/u2/m/madams/petsc_install/petsc/src/snes/examples/tutorials/ex19)
> ==3965== by 0x20104A52: kh_resize_HTPrinted (viewreg.c:12)
> ==3965== by 0x20104EE2: kh_put_HTPrinted (viewreg.c:12)
> ==3965== by 0x201061EF: PetscOptionsHelpPrintedCheck (viewreg.c:89)
> ==3965==
> ==3965== Conditional jump or move depends on uninitialised value(s)
> ==3965== at 0x27B08184: strlen (strlen.S:210)
> ==3965== by 0x200B3A48: PetscStrlen (str.c:158)
> ==3965== by 0x2011693D: PetscVSNPrintf (mprint.c:188)
> ==3965== by 0x20117006: PetscVFPrintfDefault (mprint.c:294)
> ==3965== by 0x20136BB2: PetscErrorPrintfDefault (errtrace.c:116)
> ==3965== by 0x2013938C: PetscSignalHandlerDefault (signal.c:131)
> ==3965== by 0x2013903B: PetscSignalHandler_Private (signal.c:43)
> ==3965== by 0x2335D07F: ??? (in /global/u2/m/madams/petsc_install/petsc/src/snes/examples/tutorials/ex19)
> ==3965== by 0x20104A52: kh_resize_HTPrinted (viewreg.c:12)
> ==3965== by 0x20104EE2: kh_put_HTPrinted (viewreg.c:12)
> ==3965== by 0x201061EF: PetscOptionsHelpPrintedCheck (viewreg.c:89)
> ==3965== by 0x2009AD61: PetscOptionsBegin_Private (aoptions.c:34)
> ==3965==
> ------------------------------------------------------------------------
> [0]PETSC ERROR: Caught signal number 4 Illegal instruction: Likely due to memory corruption
>
>
> On Mon, Jul 2, 2018 at 8:53 PM Mark Adams <mfadams at lbl.gov> wrote:
> Looping Treb and Baky back into the thread, and dropping PETSs.
>
> Great, thanks Barry for figuring this out.
>
> Treb and Baky need 64 bit indices, but in the mean time I can build a 32 bit version to let them test.
>
> I am all set up to test a 64 bit version. If you can give me a branch I can test.
>
> Thanks,
> Mark
>
>
>
> On Mon, Jul 2, 2018 at 7:47 PM Smith, Barry F. <bsmith at mcs.anl.gov> wrote:
>
> Progress has been made. These libraries do contain mkl_set_num_threads() is found so ./configure knows that it is MKL libraries (unlike before when it did not recognize that it was MKL libraries).
>
>
> Damn it, here is why:
>
> --with-64-bit-indices=1
>
> currently the MKL sparse stuff only works with 32 bit integers AND 32 bit integer BLAS/LAPACK.
>
> It will never work with 64 bit indices and 32 bit integer BLAS/LAPACK. but maybe could be upgraded to work with 64 bit indices and 64 bit integer BLAS/LAPACK Richard?
>
>
> Anyways the requirement in mkl_sparse.py is
>
> self.requires32bitint = 1
>
> The problem is ./configure does not print enough information to make it immediately clear it is rejecting the package because of this incompatibility.
>
>
>
> Barry
>
>
>
>
> > On Jul 2, 2018, at 5:57 PM, Mark Adams <mfadams at lbl.gov> wrote:
> >
> > Same error:
> >
> > 15:53 nid02517 master *= ~/petsc_install/petsc/src/snes/examples/tutorials$ make PETSC_DIR=/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp PETSC_ARCH="" ex19
> > cc -o ex19.o -c -g -O0 -mkl -static-intel -fopenmp -I/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/include -I/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/include -I/global/homes/m/madams/tmp/hypre-2.14.0/include -I/opt/intel/compilers_and_libraries_2018.1.163/linux/mkl/include -I/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/include `pwd`/ex19.c
> > cc -g -O0 -mkl -static-intel -fopenmp -o ex19 ex19.o -L/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/lib -Wl,-rpath,/global/homes/m/madams/tmp/hypre-2.14.0/lib -L/global/homes/m/madams/tmp/hypre-2.14.0/lib -Wl,-rpath,/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/lib -L/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/lib -L/opt/intel/compilers_and_libraries_2018.1.163/linux/mkl/lib/intel64 -lpetsc -lHYPRE -lparmetis -lmetis -lstdc++ -ldl -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread
> > /opt/intel/compilers_and_libraries_2018.1.163/linux/mkl/lib/intel64/libmkl_core.a(mkl_semaphore.o): In function `mkl_serv_load_inspector':
> > mkl_semaphore.c:(.text+0x123): warning: Using 'dlopen' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking
> > /global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp/lib/libpetsc.a(send.o): In function `PetscOpenSocket':
> > /global/u2/m/madams/petsc_install/petsc/src/sys/classes/viewer/impls/socket/send.c:108: warning: Using 'gethostbyname' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking
> > rm ex19.o
> > 15:54 nid02517 master *= ~/petsc_install/petsc/src/snes/examples/tutorials$ make PETSC_DIR=/global/homes/m/madams/petsc_install/petsc-cori-knl-dbg64-intel-omp PETSC_ARCH="" runex19_gamg
> > lid velocity = 0.0625, prandtl # = 1., grashof # = 1.
> > [0]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
> > [1]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
> > [1]PETSC ERROR: Unknown type. Check for miss-spelling or missing package: http://www.mcs.anl.gov/petsc/documentation/installation.html#external
> > [1]PETSC ERROR: Unknown Mat type given: aijmkl
> >
> > On Mon, Jul 2, 2018 at 6:43 PM Satish Balay <balay at mcs.anl.gov> wrote:
> > Hm - I suspect its an issue with mkl includes - not the libraries.
> >
> > Satish
> >
> > On Mon, 2 Jul 2018, Smith, Barry F. wrote:
> >
> > >
> > > Mark,
> > >
> > > This is not useful. We need the new configure log from when you list all the libraries NERSE recommends (which may work).
> > >
> > > I already said what was wrong with this configuration.
> > >
> > > Barry
> > >
> > >
> > > > On Jul 2, 2018, at 5:25 PM, Mark Adams <mfadams at lbl.gov> wrote:
> > > >
> > > >
> > > >
> > > > On Mon, Jul 2, 2018 at 6:22 PM Satish Balay <balay at mcs.anl.gov> wrote:
> > > > I don't understand the problem here..
> > > >
> > > > > > > [0]PETSC ERROR: Unknown type. Check for miss-spelling or missing package: http://www.mcs.anl.gov/petsc/documentation/installation.html#external
> > > > > > > [0]PETSC ERROR: Unknown Mat type given: aijmkl
> > > >
> > > > If this is the problem - then we'll have to look at configure.log to check why PETSC_HAVE_MKL_SPARSE flag is not set.
> > > >
> > > > <configure.log>
> > >
> >
> > <configure.log>
>
>
More information about the petsc-dev
mailing list