Build

 

  • Two recipes, in manage/recipes/system/hdf5-1.8.17-prod and 1.8.17-debug
  • Build issues
    • how tight a dependency to mpi, in meta.yaml?
      • currently, we require openmpi >= 1.8
      • use openmpi==1.10.3 lsf_verbs_100 for build
    • the light build dependency is convenient, but maybe a bad idea.
  • Debug build

Test

  •  install the production hdf5 into the testana environment (the same one we started on the Building Open MPI page), just do this on the rhel7dev
    • conda install hdf5=1.8.17=openmpi_100 -n testana
  • install the debug hdf in the testanadbg environment
    •  conda install hdf5=1.8.17=openmpi_dbg_100 -n testanadbg
  • as a user (like davidch), source activate testana
    • which h5ls - just verify you have new hdf5 executables
    • within the github test-lcls-conda-build-system repository, cd to hdf5
      • build and run a simple hdf5 program:
        • h5c++ h5_cmprss.c
        • ./a.out
        • do h5ls or h5dump on the output: cmprss.h5
      • testanadbg
        • In this environment, after building, run the app like
          • HD5F_DEBUG=trace,all ./a.out
          • in addition to the trace of function calls, you should see things like >deflate, <deflate that collects stats on filters
    •  Test parallel, from either, or both environments:
      • h5pcc Sample_mpio.c
      • bsub -q psanaq -I -n 20 mpirun
        • currently, this program does not succeed from the debug environment? PSRT-126 - Getting issue details... STATUS

 

 

  • No labels