[petsc-users] SUPERLU_DIST in single precision

[Barry Smith]

  On some systems, like my Mac, the environmental variable CPATH is set to the appropriate directory for mkl.h when MKL is initialized with 

source /opt/intel/compilers_and_libraries/mac/bin/compilervars.sh intel64

hence the compiler can magically find mkl.h but this doesn't seem to be universal, I think that is why Felix's system cannot find the include, I found a Linux machine at ANL that produces the same problem Felix has. Anyways I'm fixing the PETSc install will just work even if CPATH is not set.

   Barry





> On Aug 3, 2020, at 10:21 PM, Xiaoye S. Li <xsli at lbl.gov> wrote:
> 
> Regarding MKL, superlu only uses some BLAS routines in MKL.  If you have alternative BLAS, you don't need to use MKL.
> 
> But, in my experience, on an Intel system, the compiler can correctly include the path with mkl.h, that is, I don't need to do anything special. 
> I think mpiicc should work.
> 
> Sherry
> 
> On Mon, Aug 3, 2020 at 8:46 AM <flw at rzg.mpg.de <mailto:flw at rzg.mpg.de>> wrote:
> Hi Barry,
> Thanks for the branch (and thanks to Sherry as well). I tried to use  
> the configure example " arch-ci-linux-intel-mkl-single.py" (adding the  
> --with-batch flag, since I am running on a cluster), but I get the  
> following error message:
> TESTING: check from  
> config.libraries(config/BuildSystem/config/libraries.py:157)            
> 
> 
> 
> *******************************************************************************
>           UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log  
> for details):
> -------------------------------------------------------------------------------
> Downloaded ptscotch could not be used. Please check install in  
> /u/flw/petsc-hash-pkgs/073aec050
> 
> (see configure1.log)
> 
> Next I tried a minimalistic version using Intel MPI compilers:
> ./configure --download-superlu_dist --download-metis  
> --download-parmetis --download-ptscotch  --with-precision=single  
> --with-batch --with-fc=mpiifort --with-cc=mpiicc --with-cxx=mpiicpc
> 
> There I got the following error:
> ===============================================================================                                                                                                                                                                     Compiling and installing SUPERLU_DIST; this may take several minutes                                                                                                                                                                    ===============================================================================                                                                                                                                                                                                                                                                                                                                                                                                              
> *******************************************************************************
>           UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log  
> for details):
> -------------------------------------------------------------------------------
> Error running make on  SUPERLU_DIST
> 
> (see configure2.log)
> 
> 
> Next I tried
> ./configure --download-superlu_dist --download-metis  
> --download-parmetis --download-ptscotch  --with-precision=single  
> --with-batch  
> --with-mpi-dir=/mpcdf/soft/SLE_12/packages/x86_64/intel_parallel_studio/2018.4/impi/2018.4.274/intel64
> 
> Same error message (see configure3.log)
> 
> 
> It seems that there is something going on with mkl in case I want to  
> use Intel compilers for C and C++ (compiling with gcc and g++ seems to  
> work )
> 
> Do you know what is going on there?  (I am running on the Draco  
> cluster, if that helps  
> (https://www.mpcdf.mpg.de/services/computing/draco <https://www.mpcdf.mpg.de/services/computing/draco>))
> 
> 
> Best regards,
> Felix
> 
> Zitat von Barry Smith <bsmith at petsc.dev <mailto:bsmith at petsc.dev>>:
> 
> > Felix,
> >
> >     The branch is ready. Just use
> >
> >     git checkout barry/2020-07-28/superlu_dist-single
> >   ./configure --download-superlu_dist --download-metis  
> > --download-parmetis --download-ptscotch  --with-precision=single   
> > and whatever else you use
> >
> >   Barry
> >
> >   It will automatically get the needed branch of SuperLU_DIST that  
> > Sherry prepared.
> >
> >
> >> On Jul 27, 2020, at 2:10 PM, flw at rzg.mpg.de <mailto:flw at rzg.mpg.de> wrote:
> >>
> >> Hi Shery,
> >> Yes, ideally we would like to compile PETSc in single precision and  
> >> simply run a single precision version of SUPERLU_DIST just like  
> >> e.g. MUMPS.
> >>
> >> Best regards and thanks,
> >> Felix
> >> Zitat von "Xiaoye S. Li" <xsli at lbl.gov <mailto:xsli at lbl.gov>>:
> >>
> >>> Barry,
> >>>
> >>> I have a branch 'Mixed-precision' working with single precision FP32. I
> >>> assume Felix wants to use superlu_dist from petsc.  How do you want to
> >>> incorporate it in petsc?
> >>>
> >>> https://github.com/xiaoyeli/superlu_dist <https://github.com/xiaoyeli/superlu_dist>
> >>>
> >>> PS1:  in this version, FP32 only works on CPU.  FP64 and complex-FP64 all
> >>> work on GPU.
> >>>
> >>> PS2: currently there is no mixed-precision yet, but it is the branch we are
> >>> adding mix-prec support.  Will take a while before merging to master.
> >>>
> >>> Sherry
> >>>
> >>>
> >>> On Wed, Jul 22, 2020 at 6:04 AM <flw at rzg.mpg.de <mailto:flw at rzg.mpg.de>> wrote:
> >>>
> >>>> Hi Barry,
> >>>> for now I just want to run everything in single on CPUs only with
> >>>> SUPERLU_DIST. Maybe we will also incorporate GPUs in the future, but
> >>>> there are no immediate plans yet. So if you could provide the support,
> >>>> that would be awesome.
> >>>>
> >>>> Best regards,
> >>>> Felix
> >>>>
> >>>> Zitat von Barry Smith <bsmith at petsc.dev <mailto:bsmith at petsc.dev>>:
> >>>>
> >>>> > Felix,
> >>>> >
> >>>> >     What are your needs, do you want this for CPUs or for GPUs?  Do
> >>>> > you wish to run all your code in single precision or just the
> >>>> > SuperLU_Dist solver while the rest of your code double?
> >>>> >
> >>>> >     If you want to run everything on CPUs using single precision
> >>>> > then adding the support is very easy, we can provide that for you
> >>>> > any time. The other cases will require more thought.
> >>>> >
> >>>> >     Barry
> >>>> >
> >>>> >
> >>>> >> On Jul 21, 2020, at 8:58 AM, flw at rzg.mpg.de <mailto:flw at rzg.mpg.de> wrote:
> >>>> >>
> >>>> >> Dear PETSc support team,
> >>>> >> some time ago you told me that you are planning on releasing a
> >>>> >> version that supports SUPERLU_DIST in single-precision soon. Can
> >>>> >> you tell me roughly what time frame you had in mind?
> >>>> >>
> >>>> >> Best regards,
> >>>> >> Felix
> >>>> >>
> >>>>
> >>>>
> >>>>
> >>>>
> >>
> >>
> >>
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20200803/20d73eb9/attachment-0001.html>