[petsc-users] DMLocalToGlobal PETScSF is BROKEN

[Barry Smith]

  We are waiting for Jed to confirm a problem and fix it (or yell at me for unfairly blaming him).

   Now you should simply always immediately follow the Begin function call with the corresponding End function call.

   Barry

On Jan 23, 2014, at 3:42 PM, Dharmendar Reddy <dharmareddy84 at gmail.com> wrote:

> Hello, 
>           How should i fix the issue mentioned below ?
> 
> Sent from my HTC smartphone on the Now Network from Sprint!
> 
> ----- Reply message -----
> From: "Dharmendar Reddy" <dharmareddy84 at gmail.com>
> To: "Jed Brown" <jed at jedbrown.org>
> Cc: "Barry Smith" <bsmith at mcs.anl.gov>, "PETSc users list" <petsc-users at mcs.anl.gov>
> Subject: [petsc-users] DMLocalToGlobal PETScSF is BROKEN
> Date: Wed, Jan 22, 2014 5:26 pm
> 
> 
> Hello,
>       I have attached the test case which demonstrates the problem.
> 
> The following is the issue:
> globalX1=1.0
> globalX2=2.0
> 
> localX1=-1.0
> localX2=-2.0
> 
> 
> Begin: localX1 to globalX1
> Begin: localX2 to globalX2
> 
> End:   localX1 to globalX1
> End:   localX2 to globalX2
> 
> 
> i get globalX1 = -2 and globalX2 = -1.0 instead of globalX1 = -1 and
> globalX2 = -2.0
> 
> 
> When my code crashed, i was doing this to set the ub and lb vectors
> for SNES and (now) obviously the code did not progress as the ub and
> lb were swaped. Should the snes warn for such case where lb < ub is
> not valid, by user error or otherwise as in this case.
> 
> On Wed, Jan 22, 2014 at 4:39 PM, Jed Brown <jed at jedbrown.org> wrote:
> > Barry Smith <bsmith at mcs.anl.gov> writes:
> >>     Ok, well the damn code is so convoluted it is hard to see that it
> >>     is even trying to do things right for multiple independent calls.
> >
> > How would you change this to make it less convoluted?
> >
> > static PetscErrorCode PetscSFReduceEnd_Basic(PetscSF sf,MPI_Datatype unit,const void *leafdata,void *rootdata,MPI_Op op)
> > {
> >   void             (*UnpackOp)(PetscInt,const PetscInt*,void*,const void*);
> >   PetscErrorCode   ierr;
> >   PetscSFBasicPack link;
> >   PetscInt         i,nrootranks;
> >   const PetscInt   *rootoffset,*rootloc;
> >
> >   PetscFunctionBegin;
> >   ierr = PetscSFBasicGetPackInUse(sf,unit,rootdata,PETSC_OWN_POINTER,&link);CHKERRQ(ierr);
> >   /* This implementation could be changed to unpack as receives arrive, at the cost of non-determinism */
> >   ierr = PetscSFBasicPackWaitall(sf,link);CHKERRQ(ierr);
> >   ierr = PetscSFBasicGetRootInfo(sf,&nrootranks,NULL,&rootoffset,&rootloc);CHKERRQ(ierr);
> >   ierr = PetscSFBasicPackGetUnpackOp(sf,link,op,&UnpackOp);CHKERRQ(ierr);
> >   for (i=0; i<nrootranks; i++) {
> >     PetscMPIInt n          = rootoffset[i+1] - rootoffset[i];
> >     const void  *packstart = link->root+rootoffset[i]*link->unitbytes;
> >
> >     (*UnpackOp)(n,rootloc+rootoffset[i],rootdata,packstart);
> >   }
> >   ierr = PetscSFBasicReclaimPack(sf,&link);CHKERRQ(ierr);
> >   PetscFunctionReturn(0);
> > }
> >
> >
> >>     You’ll have to pry out of Dhamendar his test case that fails.
> >
> > That would be useful.
> 
> 
> 
> -- 
> -----------------------------------------------------
> Dharmendar Reddy Palle
> Graduate Student
> Microelectronics Research center,
> University of Texas at Austin,
> 10100 Burnet Road, Bldg. 160
> MER 2.608F, TX 78758-4445
> e-mail: dharmareddy84 at gmail.com
> Phone: +1-512-350-9082
> United States of America.
> Homepage: https://webspace.utexas.edu/~dpr342