Re: [mpich-devel] MPIR_ERRTEST_ARGNULL and count=0 (or type_size=0)
I don't know if you read http://trac.mpich.org/projects/mpich/ticket/1863, but the RMA code in the repo is using MPIR_ERRTEST_ARGNULL in a few places, which caused me problems in ARMCI-MPI. Most of the RMA functions, in particular the ones that existed in MPI-2, were missing any buffer checks. The buffer checks that were present were added to the RMW routines. My patch is relative to remotes/origin/HEAD. "I don't always build MPI from source, but when I do, I use the latest development version of MPICH" </dos equis mem> Best, Jeff On Wed, May 29, 2013 at 4:14 PM, Lisandro Dalcin <[email protected]> wrote:
On 29 May 2013 23:55, Jeff Hammond <[email protected]> wrote:
Well that's no fun. I was hoping to make it look like I did some actual work here :-)
The actual work is figuring out the issue!!! I've spent enough time with these bugs in other MPI implementations. As I know by testing myself that MPICH does the right thing for MPI 2.x calls, I guesses you were missing a more easy way to do it.
BTW, check this:
$ grep -r MPIR_ERRTEST_USERBUFFER src/mpi/rma
I get no output, I think other RMA calls also need better error checking of buffer addresses...
-- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169
-- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute [email protected] / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond ALCF docs: http://www.alcf.anl.gov/user-guides
participants (1)
-
Jeff Hammond