problems with ACML and single precision

[Roberto Gori]

Hi Satish,

i would use C. 
I mean ... for the single precision i would like to use the acml library with the intel  C compiler as in petsc 2.3.1.
Is this impossibile for the current version or in the next future?

    Roberto


----- Original Message ----- 
  From: Satish Balay 
  To: petsc-users at mcs.anl.gov 
  Sent: Friday, June 29, 2007 3:28 PM
  Subject: Re: problems with ACML and single precision


  yeah - singleprecision requires cblaslapack, and cblaslapack can't be
  used in conjunction with fortran.

  Are you planning to use single precision PETSc from fortran? If so,
  currently its not possible. [we'll have to figureout how to get this
  working].

  However if you wish do this from c, then just use the additional
  configure option --with-fc=0

  Satish

  On Fri, 29 Jun 2007, Roberto Gori wrote:

  > Thank you Matthew,
  > 
  > I tried,
  > 
  > now my configure looks like this:
  > 
  > >config/configure.py --with-mpirun=mpirun.lsf --with-shared=0  --with-precision=single --download-c-blas-lapack=yes
  > 
  > and I get the following error:
  > 
  > =================================================================================
  >              Configuring PETSc to compile on your system                         
  > =================================================================================
  > TESTING: configureLibrary from config.packages.BlasLapack(python/BuildSystem/config/packages/BlasLapack.py:384)                                                                               *********************************************************************************
  >          UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log for details):
  > ---------------------------------------------------------------------------------------
  > Should request f-blas-lapack, not --download-c-blas-lapack=yes since you have a fortran compiler?
  > *********************************************************************************
  > 
  > so i changed my configure:
  > 
  > >config/configure.py --with-mpirun=mpirun.lsf --with-shared=0  --with-precision=single --download-f-blas-lapack=yes 
  > 
  > =================================================================================
  >              Configuring PETSc to compile on your system                         
  > =================================================================================
  > =================================================================================                                                                                                                   
  > Compiling FBLASLAPACK; this may take several minutes                                                                                                                                    =================================================================================                                                                                                             
  > TESTING: checkMissing from config.packages.BlasLapack(python/BuildSystem/config/packages/BlasLapack.py:446)                                                                                   *********************************************************************************
  >          UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log for details):
  > ---------------------------------------------------------------------------------------
  > Need to use --download-c-blas-lapack when using --with-precision=longdouble/single
  > *********************************************************************************
  > 
  > 
  > I'm in loop.
  > 
  > Moreover to achieve better performance i would prefer to use acml instead of the BlasLapack sources.
  > Everything works when i build petsc 2.3.1 with the Intel's compiler. I'm not able to do that with petsc 2.3.2.
  > 
  >     Roberto
  >   ----- Original Message ----- 
  >   From: Matthew Knepley 
  >   To: petsc-users at mcs.anl.gov 
  >   Sent: Thursday, June 28, 2007 8:50 PM
  >   Subject: Re: problems with ACML and single precision
  > 
  > 
  >   Yes, you need to let us compile for single precision. Use the option in the
  >   error message.
  > 
  >     Thanks,
  > 
  >       Matt
  > 
  >   On 6/28/07, Roberto Gori <r.gori at cineca.it> wrote:
  >   >
  >   >
  >   > Hi,
  >   >
  >   > i'm trying to install PETSc 2.3.3-p3 on a Linux amd64 cluster.
  >   >
  >   > There's no problems wìth double precison but when I add the single precision
  >   > option to my configure command I get the following error:
  >   >
  >   > =================================================================================
  >   >              Configuring PETSc to compile on your system
  >   >
  >   > =================================================================================
  >   > TESTING: checkMissing from
  >   > config.packages.BlasLapack(python/BuildSystem/config/packages/BlasLapack.py:446)
  >   >
  >   >
  >   > *********************************************************************************
  >   >          UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log for
  >   > details):
  >   > ---------------------------------------------------------------------------------------
  >   > Need to use --download-c-blas-lapack when using
  >   > --with-precision=longdouble/single
  >   > *********************************************************************************
  >   >
  >   >
  >   > This is my configure:
  >   > config/configure.py --with-mpirun=mpirun.lsf --with-shared=0
  >   > --with-blas-lib=/cineca/prod/acml/3.6.0/ifort64/lib/libacml.a
  >   > --with-lapack-lib=/cineca/prod/acml/3.6.0/ifort64/lib/libacml.a
  >   > --with-precision=single
  >   >
  >   >
  >   > Maybe ACML is not supported for single precision?
  >   >
  >   > Thanks
  >   >
  >   >     Roberto
  > 
  > 
  >   -- 
  >   What most experimenters take for granted before they begin their
  >   experiments is infinitely more interesting than any results to which
  >   their experiments lead.
  >   -- Norbert Wiener
  > 
  > 
  >
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20070629/6ff2e136/attachment.htm>