Boyce: I recognize the benefits and we should talk about this more. In my case I do have my own data types too for interlaced data and it would be nice to have it all work well. As you know, it's non trivial maintaining all this.
The reason this whole conversation started was difficulties in trying to keep PFLOTRAN, SAMRAI, and PETSc in sync. Bobby ________________________________________ From: Boyce Griffith [griff...@cims.nyu.edu] Sent: Tuesday, June 07, 2011 1:05 PM To: For users of the development version of PETSc Cc: Philip, Bobby Subject: Re: [petsc-dev] Integrating PFLOTRAN, PETSC & SAMRAI On 6/7/11 12:48 PM, Bobby Philip wrote: > > On Jun 6, 2011, at 4:26 PM, Boyce Griffith wrote: >> >> What I'd really like to do is to setup some SAMRAI stuff that actually >> uses Vec / Mat for data storage, so that most of this wrapper stuff is >> not necessary. (It would probably still be necessary to have a wrapper >> for VecNorm, but that might be the only one.) Of course, I've been >> wanting to do this for years now... > > Setting it up to use PETSc storage would probably break a lot of my code and > PFLOTRAN. Not that I have the time to do this right now, but the idea would be to do most/all of this "behind the scenes," so that application codes that use standard SAMRAI PatchData types would work without modification. One benefit of doing this could be easier access to different smoothers for use in FAC; or to other composite-grid solvers for problems where FAC is not necessary. Another might be easier access to CUDA. -- Boyce
