On 4 June 2009 at 21:16, Dirk Eddelbuettel wrote: | On 18 May 2009 at 14:41, Adam C Powell IV wrote: | | On Mon, 2009-05-18 at 13:25 -0500, Dirk Eddelbuettel wrote: | | > On 18 May 2009 at 19:57, Christophe Prud'homme wrote: | | > | Adam | | > | | | > | here is what I get at link time with the current (3.0) petsc version | | > | | | > | /usr/lib/libpetsc.a(sprng.o): In function | | > | `PetscRandomGetValueImaginary_Sprng': | | > | (.text+0xe1): undefined reference to `get_rn_dbl_simple_mpi' | | > | /usr/lib/libpetsc.a(sprng.o): In function | | > | `PetscRandomGetValueImaginary_Sprng': | | > | (.text+0xfa): undefined reference to `get_rn_dbl_simple_mpi' | | > | /usr/lib/libpetsc.a(sprng.o): In function `PetscRandomGetValueReal_Sprng': | | > | (.text+0x142): undefined reference to `get_rn_dbl_simple_mpi' | | > | /usr/lib/libpetsc.a(sprng.o): In function `PetscRandomGetValueReal_Sprng': | | > | (.text+0x15b): undefined reference to `get_rn_dbl_simple_mpi' | | > | /usr/lib/libpetsc.a(sprng.o): In function `PetscRandomGetValue_Sprng': | | > | (.text+0x1a3): undefined reference to `get_rn_dbl_simple_mpi' | | > | /usr/lib/libpetsc.a(sprng.o):(.text+0x1bc): more undefined references to | | > | `get_rn_dbl_simple_mpi' follow | | > | /usr/lib/libpetsc.a(sprng.o): In function `PetscRandomSeed_Sprng': | | > | (.text+0x1ef): undefined reference to `init_rng_simple_mpi' | | > | | | > | Given Dirk's comment regarding sprng, perhaps we should disable it for now ? | | | | Yup, it's disabled in my current build. Will upload when I finish | | testing. | | | | > So is the following a concise summary | | > | | > sprng2 available in debian, and used by r-cran-rsprng | | | | But built without MPI. | | | | > sprng4 _not yet_ available in debian, but possibly useful | | > for PetSc ? | | > | | > I.e. no overlap? As I understand the sprong website, versions 2 and 4 are | | > not plug-and-play. So if you guys have use of for sprng _version 4_ you may | | > just have to package it, no? I could then try to port the R package | | > forward and we could all get rid of the old sprng2 | | | | Actually, PETSc's linkage is made for version 1 (made some changes to | | try to support version 2), so we shouldn't be a factor in this decision. | | I wanted to do some RNG stuff and added the link, but it seems that was | | misguided. | | What is your platform? I just build a test version of libsprng2 where I | enable MPI. It does not seem to affect the use by R --- on an example init I | get the same sequence on RNGs with the prior (non-MPI) build and the current | one. Good. | | I'd love to get you that for testing. It's a trivial build in a chroot or | directly, and incase you use x86 (rather than x86-64 as I suspect/recall you | do) I could send you the lib and -dev packages.
I did not hear from anyone, so I can only conclude that the interest in an mpi-enhanced sprng is not all that strong. An additional problem is the arane build process. We'd have to hardcode paths which, given our open mpi and lam split over different architectures, essentially means doing everything twice. I don't think I want to do that. So I'd like to close this bug for lack of relevance. Voice your opposition now ... else I move ahead in a few days and close this. Dirk | Dirk | | -- | Three out of two people have difficulties with fractions. -- Three out of two people have difficulties with fractions. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org