######################################
# CG UPC Benchmarks                  #
# Revision 1.1        06/26/2001     #
#                                    #
# Department of Computer Science     #
# University of North Carolina       #
#                                    #
# email: {feasel, huan, prins} @     # 
#         cs.unc.edu                 #
# http://www.cs.unc.edu/~prins       #
######################################

==========================================
INSTALLATION
    
    Three UPC CG implementations were supported
    in this package. They were named as cgB.c(upc)
    cgL1.c(upc) and cgL2.c(upc). Please refer to the 
    associated report for the used parallel schema.

    To install the bench mark, 
    1). Create a subdirectory bin under UPC subdirectory
    2). Choosing the right configuration file in config 
        subdirectory. Two platforms are supported in 
        this package: SGI origin 2000 and Compaq Alpha cluster.
        One configuration template is provided for each of the 
        two platforms. 
    3). run the make file with the following parameters:
        >make <benchmark-name> CLASS=<class> THREADS=<threads number>
        where <benchmark-name> := "CG1" | "CG2" | "CG3" 
              <class>          := "S" | "W" | "A" | "B" | "C"
              <threads number> := a nature number
 
    The executable is put in the bin directory
    with the format:
       cg<suffix>.<class>.<threads number>
       where <suffix> := "B" | "L1" | "L2"
    
    To run the benchmark on alpha sever, use prun 
    to allocate parallel processes 
      >prun -n <threads number> ./<executable>
    
    To run the benchmark on SGI origin2000
      ><executable> -fupc-threads-<THREADS>

    Note: cleaning all the .o and the executable file
    setparams (under sys/) is recommended when switching machines
    which are running under the same OS. It is required if 
    switching between different operating systems. 
 
Change since 1.0
==================================================================
    The last two  UPC implementations (L1 and L2) in 1.0
    is combined into on program (L1) in 1.1. A new implementation,
    which follows the MPI scheme and supports the 2D decomposition
    of the matrix is provided in 1.1 (L2). 
  
    SGI origin 2000 support is made available in this release. This
    is done by 1). providing two identical source files 
    with different suffix. This is used to satisfy the different 
    compiler requirement.  2). providing two configuration templates 
    for the two platforms (under config/). Please use the right 
    template according to the execution platforms. 
    
    Limited code change to L1 was made in order to 
    improve the readability and a bug reported in 
    the makefile was fixed in this release.
    
==========================================
Acknowledgement:
    The MPI implementation is available from NASA website at 
         http://www.nas.nasa.gov/NAS/NPB/           
    The OpenMP implementation is available on line at
         http://pdplab.trc.rwcp.or.jp/pdperf/Omni/benchmarks/NPB/
  

Send comments or suggestions to  prins@cs.unc.edu