Omnia vincit amor
Home -> Research -> NB Collectives -> HPL
Home
  Publications
  Awards
  Research
    
NB Collectives
      
LibNBC
      NBCBench
      Performance
      CG Solver
      Kernels
      HPL
    MPI Topologies
    MPI Datatypes
    Netgauge
    Network Topologies
    Ethernet BTL eth
    ORCS
    DFSSSP
    Older Projects
    cDAG
    LogGOPSim
    CoMPIler
  Teaching
  Miscellaneous
  Full CV [pdf]
  BLOG






  Events








  Past Events





An Example: High Performance Linpack

This is a patch to the original High-Performance Linpack Benchmark (HPL Version 1.0a) that enables the usage of non-blocking collective operations. This patch is based on Christian Siebert's "native MPI_Bcast() patch" for HPL and adds another broadcast option that lets the user select non-blocking collective operation implemented in LibNBC. Applying the patch (pseudo-script):

 mkdir linpack
 cd linpack
 # download original HPL
 wget http://www.netlib.org/benchmark/hpl/hpl.tgz
 # unpack HPL
 tar xzf hpl.tgz
 # download patch 
 wget http://www.unixer.de/research/nbcoll/hpl/hpl-1.0a-NBC.patch
 cd hpl
 # patch HPL
 patch -p 1 < ../hpl-1.0a-NBC.patch
 # create Make.Linux according to HPL documentation
 # add LibNBC include (nbc.h) to CFLAGS
 # add LibNBC library (libnbc.a) to LDFLAGS
 # compile
 # create HPL.dat (select 6 or 7 for Bcast) and run 


The patch is available here: Patch against HPL-1.0a to enable MPI_Bcast() and NBC_Ibcast() - (22.99 kb)
#########################################################################

file    hpl-1.0a-NBC.patch
for     a "MPI_Bcast() and NBC_Ibcast() patch" for the 
        High-Performance Linpack Benchmark
by      Torsten Hoefler, Christian Siebert

#########################################################################

serving: 18.97.14.83:57894© Torsten Hoefler