Tested by IBM
  Portability Flags:
    -qfixed used in:        168.wupwise, 171.swim, 172.mgrid, 173.applu,
                            178.galgel, 200.sixtrack, 301.apsi
    -qsuffix=f=f90 used in: 178.galgel, 187.facerec, 189.lucas, 191.fma3d


  Base Optimization Flags:
    C:         -qpdf1/pdf2
               -O5 -blpdata -lmass
    Fortran:   -qpdf1/pdf2
               -O5 -blpdata -lmass

  Peak Optimization Flags:
     168.wupwise:    -O5
     171.swim:       basepeak=1
     172.mgrid:      -qpdf1/pdf2
                     -O5 -blpdata -lmass
                     users=32
     173.applu:      -O5 -qarch=pwr3 -qtune=pwr3 -blpdata -lmass
                     users=32
     177.mesa:       -qpdf1/pdf2
                     -O5
     178.galgel:     -O5 -blpdata -qessl -lessl
     179.art:        -O5 -lmass -qessl -lessl -blpdata -qsave
                     users=32
     183.equake:     -qpdf1/pdf2
                     -O5 -blpdata -lmass
                     users=32
     187.facerec:    -O3 -qhot -qarch=pwr5 -qtune=pwr5 -qfdpr
                     fdpr -R3
     188.ammp:       -qpdf1/pdf2
                     -O5 -blpdata -qalign=natural -D_ILS_MACROS
     189.lucas:      -O5 -blpdata -lmass
     191.fma3d:      -qpdf1/pdf2
                     -O5 -blpdata -qalign=natural -qhot=arraypad -Q
     200.sixtrack:   -O3 -qhot -qarch=pwr5 -qtune=pwr5 -qfdpr
                     fdpr -R3
     301.apsi:       -O5 -lmass -qessl -lessl -blpdata -qsave


  SMT: Acronym for "Simultaneous Multi-Threading". A processor technology that allows
       the simultaneous execution of multiple thread contexts within a single processor
       core. (Enabled by default)
  MCM: Acronym for "Multi-Chip Module" (four dual-core processor chips + four L3-cache chips)
       This system contains 4 MCMs.
  SUT: Acronym for "System Under Test"
  ESSL: Engineering and Scientific Subroutine Library

  C:       IBM XL C for AIX invoked as xlc
  Fortran: IBM XL Fortran for AIX invoked as xlf90

  APAR IY60349 was applied to AIX to enable new hardware support.
  ulimits set to unlimited.
  Large page mode and memory affinity were set as follows:
     vmo -r -o lgpg_regions=4096 -o lgpg_size=16777216 -o memory_affinity=1
     chuser capabilities=CAP_BYPASS_RAC_VMM,CAP_PROPAGATE $USER
     shutdown -r
     export MEMORY_AFFINITY=MCM

  The following config-file entry was used to assign each benchmark process to a core:
     submit = schedule.64 \$SPECUSERNUM $command
  with the "schedule.64" function defined as follows:
     #!/bin/ksh
     index=$1
     shift 1                         # Strip off the residual arguments; the rest is the command.
     if [[ $index -ge 32 ]]
     then
         target=$((1+2*(index-32)))
     else
         target=$((2*index))
     fi
     bindprocessor $$ $target        # Schedule this job to the corresponding core.
     $*                              # Now run the command.

  The "bindprocessor" AIX command binds a process to a CPU core.