From: Ben Allan ([email protected])
Date: Wed May 26 2004 - 09:16:23 PDT
Hi, For berkeley Version 1.1.0 I have upc installation/test troubles. First about my platform, so someone might be able to say 'yeah, that's not *supposed*' to work if that is the case. (If 'not' is the case, is anyone interested in this platform enough to help me out in porting?) Platform: gcc (GCC) 3.2 20020903 (Red Hat Linux 8.0 3.2-7) Linux cluster (128 nodes of dual xeon mb) (redhat 8 more or less) w/mellanox infiniband, LA-MPI (not to be confused with lam) running over infiniband, and BPROC for managing the nodes/jobs. At configure time, although we set up the environment vars to be consistent with how we run jobs using LAMPI, setenv MPI_CC /usr/bin/gcc setenv MPI_CFLAGS "-I/usr/local/lampi/include" setenv MPI_LIBS /usr/local/lampi/lib/libmpi.a setenv MPIRUN_CMD "/usr/local/lampi/bin/mpirun -n 1 -dev ib" or in case upc tools are supposed to supply a number %N we also tried: setenv MPIRUN_CMD "/usr/local/lampi/bin/mpirun -n %N -dev ib" we get "checking for working MPI configuration... no" out of the gasnet configure which is invoked from the higher level configure. The script appears to be doing some sort of link-only test, so i'm not clear why it should fail given the env vars. Is there likely to be anything in 'stable' release that makes it a better candidate for working? I didn't see much in the readmes/changelogs. Is anyone else out there looking at upc/gasnet on an infiniband/bproc/lampi cluster? thanks, Ben Allan sandia national labs livermore, ca