# Invocation command line:
# /data1/SPECcpu2kv1.3/bin/runspec -c 256lw-sles9-psc-pgi.cfg -T base,peak --reportable fp
############################################################################
############################################################################
# SPEC CPU2000 SLES9 AMD64 config file
# March 2006 Edition. 
# PathScale EKOPath Compiler Suite 2.3
# The Portland Group Compiler version 6.1-3  

action      = validate
ext=u45-sles9-psc-pgi
tune        = base,peak
PATHSEP     = /

check_md5 	=1
mean_anyway	=1
output_format  	=asc,config,html
reportable	=1

############## uncomment for RATE runs #########################
#default=default=default=default:
#submit= MYMASK=\$((1<<\$SPECUSERNUM)) ; /usr/bin/taskset \$MYMASK $command
############## uncomment for RATE runs #########################

################################################################
# Compilers
################################################################

default=default=default=default:
ACML_PSC64_DIR = /opt/acml3.0.0/pathscale64/lib
CC=pathcc
FC=pathf95
F77=pathf95
CXX=pathCC

################################################################
# System information
# If some remarks about BIOS or Firmware are needed, place them here.
################################################################

default=default=default=default:
notes900=
notes901=Default BIOS settings was used.

notes905=taskset has been used to bind processes to CPUs
notes999=

################################################################
# Description Hardware and Software
################################################################
default=default=default=default:
hw_vendor     = Sun Microsystems
tester_name   = Sun Microsystems, Santa Clara
company_name  = Sun Microsystems
hw_model     = Sun Fire X4100/X4200
hw_cpu       = AMD Opteron (TM) 256
hw_cpu_mhz   = 3000
hw_disk      = SAS,36GB,7200 RPM
hw_fpu       = Integrated
hw_memory    = 16GB (8x2GB, PC3200 CL3 DDR ECC Registered SDRAM)
hw_vendor    = Sun Microsystems 
hw_avail     = Apr-2006
hw_ncpu      = 2 cores, 2 chips, 1 core/chip
hw_ncpuorder = 1,2 (order by number of chips)
hw_ocache    = N/A
hw_other     = None
hw_parallel  = No
hw_pcache    = 64KBI + 64KBD (on chip) per core
hw_scache    = 1024KB (I+D) (on chip) per core
hw_tcache    = N/A
sw_file      = ufs
sw_os        = SuSE SLES9 with SP3
sw_state     = Multi-user
prepared_by  = Gnanakumar Rajaram (Rajaram.Gnanakumar@SUN.COM)
hw_vendor    = Sun Microsystems
tester_name  = Sun Microsystems, Santa Clara
license_num  = 6
test_date    = Mar-2006
sw_avail     = Mar-2006
sw_compiler1 = PathScale EKOPath(TM) Compiler Suite, Version 2.3
sw_compiler2 = PGI Compiler for Linux, Release 6.1
sw_compiler3 = AMD Core Mathematical Library (ACML), Version 3.0.0

# ---- End inclusion of '/home/work/CPU2000/config/SUT.inc'



################################################################
# Portability
################################################################

default=default=default=default:
notes020=Portability flags:

186.crafty=default=default=default:
CPORTABILITY = -DLINUX_i386
notes040= 186.crafty:  -DLINUX_i386

252.eon=default=default=default:
CXXPORTABILITY = -DHAS_ERRLIST -DSPEC_CPU2000_LP64
notes050= 252.eon:     -DHAS_ERRLIST -DSPEC_CPU2000_LP64

253.perlbmk=default=default=default:
CPORTABILITY = -DSPEC_CPU2000_LINUX_I386 -DSPEC_CPU2000_NEED_BOOL -DSPEC_CPU2000_GLIBC22 -DSPEC_CPU2000_LP64
notes060= 253.perlbmk: -DSPEC_CPU2000_LINUX_I386 -DSPEC_CPU2000_NEED_BOOL
notes061=              -DSPEC_CPU2000_GLIBC22 -DSPEC_CPU2000_LP64

254.gap=default=default=default:
CPORTABILITY=-DSYS_IS_USG -DSYS_HAS_IOCTL_PROTO -DSYS_HAS_TIME_PROTO -DSYS_HAS_CALLOC_PROTO -DSPEC_CPU2000_LP64
notes070= 254.gap:     -DSYS_IS_USG -DSYS_HAS_IOCTL_PROTO -DSYS_HAS_TIME_PROTO
notes071=              -DSYS_HAS_CALLOC_PROTO -DSPEC_CPU2000_LP64

255.vortex=default=default=default:
CPORTABILITY=-DSPEC_CPU2000_LP64
notes080= 255.vortex:  -DSPEC_CPU2000_LP64

178.galgel=base=default=default:
FPORTABILITY= -Mfixed
notes030=   178.galgel (base using pgf90)   : -Mfixed

178.galgel=peak=default=default:
FPORTABILITY=-fixedform
notes031=   178.galgel (peak using pathf95) : -fixedform

################################################################
# For feedback optimization; ACML-Library
################################################################

default=default=default=default:
notes100=
notes101=Feedback Optimization:
notes102=+FDO: 
notes103=       PGI     : PASS1=-Mpfi               PASS2=-Mpfo
notes104=       PathSale: PASS1=-fb_create fbdata   PASS2=-fb_opt fbdata

178.galgel=peak=default=default:
notes109=
notes111=+ACML  means -L<acml-install-dir>/pathscale64/lib -lacml,
notes112=        which links with AMD Core Math Library

################################################################
# Baseline Tuning Flags
################################################################

########
#  int
########

int=base=default=default:
PASS1_CFLAGS  = -fb_create fbdata
PASS1_CXXFLAGS  = -fb_create fbdata
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_CXXFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata

COPTIMIZE      = -Ofast
CXXOPTIMIZE    = -Ofast

notes150=
notes151=Baseline Optimization Flags:
notes152=   C programs:       -Ofast +FDO
notes153=   C++ programs:     -Ofast +FDO

########
#  fp
########

fp=base=default=default:
CC      = pgcc
FC      = pgf90
F77     = pgf90
FOPTIMIZE      = -fastsse -Mipa=fast,inline
F77OPTIMIZE    = -fastsse -Mipa=fast,inline
COPTIMIZE      = -fastsse -Mipa=fast,inline
PASS1_CFLAGS   = -Mpfi
PASS2_CFLAGS   = -Mpfo
PASS1_FFLAGS   = -Mpfi
PASS2_FFLAGS   = -Mpfo
PASS1_F77FLAGS = -Mpfi
PASS2_F77FLAGS = -Mpfo
ONESTEP=yes

#  Pathscale base settings
#COPTIMIZE   = -Ofast -WOPT:mem_opnds=on
#FOPTIMIZE   = -Ofast -LNO:fusion=2 -OPT:fast_complex=on
#F77OPTIMIZE = -Ofast -LNO:fusion=2 -OPT:fast_complex=on

notes150=
notes151=Baseline Optimization Flags:
notes152=   C programs      : pgcc  -fastsse -Mipa=fast,inline +FDO
notes153=   Fortran programs: pgf90 -fastsse -Mipa=fast,inline +FDO

################################################################
# Peak Tuning Flags
################################################################

default=peak=default=default:
notes200=
notes201=Peak Tuning Flags:

########
#  int
########

#  Feedback Optinization: Default for all int

int=peak=default=default:
PASS1_CFLAGS  = -fb_create fbdata
PASS1_CXXFLAGS  = -fb_create fbdata
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_CXXFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata

164.gzip=peak=default=default:
OPTIMIZE = -O3 -ipa -WOPT:val=0 -OPT:unroll_size=0
notes210= 164.gzip:   -O3 -ipa -WOPT:val=0 -OPT:unroll_size=0 +FDO

175.vpr=peak=default=default:
OPTIMIZE = -O3 -ipa -m32
notes220= 175.vpr:    -O3 -ipa -m32 +FDO

176.gcc=peak=default=default:
OPTIMIZE      = -O3 -IPA:plimit=10000 -LNO:opt=0 -OPT:goto=off
notes230= 176.gcc:    -O3 -IPA:plimit=10000 -LNO:opt=0 -OPT:goto=off +FDO

181.mcf=peak=default=default:
OPTIMIZE      = -O3 -ipa -IPA:field_reorder=on -m32
notes240= 181.mcf:    -O3 -ipa -IPA:field_reorder=on -m32 +FDO

186.crafty=peak=default=default:
OPTIMIZE      = -Ofast -CG:local_fwd_sched=on -LNO:opt=0 -WOPT:val=0
notes250= 186.crafty: -Ofast -CG:local_fwd_sched=on -LNO:opt=0 -WOPT:val=0 +FDO

197.parser=peak=default=default:
OPTIMIZE      = -O3 -ipa -m32 -IPA:ctype=on
notes260= 197.parser: -O3 -ipa -m32 -IPA:ctype=on +FDO

252.eon=peak=default=default:
OPTIMIZE = -Ofast -CG:gcm=off:p2align_freq=1:prefetch=off -IPA:plimit=4000 -OPT:treeheight=on -TENV:X=4:frame_pointer=off -fno-exceptions -LNO:fu=10:full_unroll_outer=on -GRA:optimize_boundary=on
notes270= 252.eon:    -Ofast -CG:gcm=off:p2align_freq=1:prefetch=off -IPA:plimit=4000
notes271=             -OPT:treeheight=on -TENV:X=4:frame_pointer=off -fno-exceptions
notes272=             -LNO:fu=10:full_unroll_outer=on -GRA:optimize_boundary=on +FDO

253.perlbmk=peak=default=default:
OPTIMIZE = -O2 -ipa -OPT:Ofast:transform_to_memlib=off -fno-math-errno -IPA:plimit=10000
notes280= 253.perlbmk:-O2 -ipa -OPT:Ofast:transform_to_memlib=off
notes281=             -fno-math-errno -IPA:plimit=10000 +FDO

254.gap=peak=default=default:
OPTIMIZE = -Ofast -IPA:min_hotness=5 -OPT:alias=disjoint -CG:movnti=1
notes290= 254.gap:    -Ofast -IPA:min_hotness=5 -OPT:alias=disjoint -CG:movnti=1 +FDO

255.vortex=peak=default=default:
OPTIMIZE = -Ofast -OPT:goto=off
notes300= 255.vortex: -Ofast -OPT:goto=off +FDO

256.bzip2=peak=default=default:
basepeak = true
notes310= 256.bzip2:  basepeak = true

300.twolf=peak=default=default:
OPTIMIZE      = -O2 -CG:gcm=off:p2align_freq=100000 -OPT:Ofast:unroll_times_max=8:unroll_size=256:alias=disjoint -WOPT:mem_opnds=on -m32
notes320= 300.twolf:  -O2 -CG:gcm=off:p2align_freq=100000
notes321=             -OPT:Ofast:unroll_times_max=8:unroll_size=256:alias=disjoint
notes322=             -WOPT:mem_opnds=on -m32 +FDO

########
#  fp
########

#  Feedback Optinization: Default for all fp

fp=peak=default=default:
PASS1_CFLAGS   = -fb_create fbdata
PASS1_FFLAGS   = -fb_create fbdata
PASS1_F77FLAGS = -fb_create fbdata
PASS1_LDFLAGS  = -fb_create fbdata
PASS2_CFLAGS   = -fb_opt fbdata
PASS2_FFLAGS   = -fb_opt fbdata
PASS2_F77FLAGS = -fb_opt fbdata
PASS2_LDFLAGS  = -fb_opt fbdata

168.wupwise=peak=default=default:
F77OPTIMIZE = -Ofast -LNO:prefetch_ahead=5:prefetch=3 -OPT:unroll_times_max=8:unroll_size=128:IEEE_NaN_Inf=off:ro=3 -IPA:linear=on:plimit=50000:callee_limit=5000 -CG:local_fwd_sched=on -m3dnow
feedback = 0
notes210= 168.wupwise:  pathf95 -Ofast -LNO:prefetch_ahead=5:prefetch=3
notes211=               -OPT:unroll_times_max=8:unroll_size=128:IEEE_NaN_Inf=off:ro=3
notes212=               -IPA:linear=on:plimit=50000:callee_limit=5000
notes213=               -CG:local_fwd_sched=on -m3dnow

171.swim=peak=default=default:
F77OPTIMIZE     = -Ofast -CG:local_fwd_sched=on -LNO:fusion=2 -m3dnow
feedback = 0
notes220= 171.swim:     pathf95 -Ofast -CG:local_fwd_sched=on -LNO:fusion=2 -m3dnow

172.mgrid=peak=default=default:
F77OPTIMIZE = -Ofast -CG:gcm=off -OPT:IEEE_a=3:unroll_size=200 -LNO:fusion=2:fission=1:blocking=off:prefetch_ahead=2 -WOPT:mem_opnds=on:aggstr=0
feedback = 0
notes230= 172.mgrid:    pathf95 -Ofast -CG:gcm=off -OPT:IEEE_a=3:unroll_size=200
notes231=               -LNO:fusion=2:fission=1:blocking=off:prefetch_ahead=2
notes232=               -WOPT:mem_opnds=on:aggstr=0

173.applu=peak=default=default:
F77OPTIMIZE = -Ofast -CG:local_fwd_sched=on -OPT:ro=3 -TENV:X=3 -LNO:fusion=2:fission=2:full_unroll_size=10000:prefetch=3
notes240= 173.applu:    pathf95 -Ofast -CG:local_fwd_sched=on -OPT:ro=3 -TENV:X=3
notes241=               -LNO:fusion=2:fission=2:full_unroll_size=10000:prefetch=3
notes242=               +FDO

177.mesa=peak=default=default:
COPTIMIZE = -O2 -ipa -OPT:Ofast -fno-math-errno -CG:local_fwd_sched=on -WOPT:mem_opnds=on
notes250= 177.mesa:     pathf95 -O2 -ipa -OPT:Ofast -fno-math-errno -CG:local_fwd_sched=on
notes251=               -WOPT:mem_opnds=on +FDO

178.galgel=peak=default=default:
FOPTIMIZE = -Ofast -OPT:fast_complex=on
RM_SOURCES      = lapak.f90
EXTRA_LIBS      = -L$(ACML_PSC64_DIR) -lacml
notes260= 178.galgel:   -Ofast -OPT:fast_complex=on +ACML +FDO

179.art=peak=default=default:
CC      = pgcc
COPTIMIZE= -fastsse -Munroll=n:9 -Mipa=fast,inline -tp k8-32
feedback = 0
notes270= 179.art:      pgcc -fastsse -Munroll=n:9 -Mipa=fast,inline -tp k8-32

183.equake=peak=default=default:
CC      = pgcc
COPTIMIZE= -fastsse -Mflushz -Mnovect -Mipa=fast,inline
PASS1_CFLAGS    = -Mpfi
PASS2_CFLAGS    = -Mpfo
# To override fp=peak settings
PASS1_LDFLAGS   = 
PASS2_LDFLAGS   = 
notes280= 183.equake:   pgcc -fastsse -Mflushz -Mnovect -Mipa=fast,inline ONESTEP=yes +FDO 
ONESTEP=yes

187.facerec=peak=default=default:
basepeak=1
notes290= 187.facerec:  pgf90  basepeak=1

188.ammp=peak=default=default:
COPTIMIZE = -O3 -OPT:alias=disjoint:unroll_times_max=8:Ofast:ro=3 -fno-math-errno -TENV:X=4
notes300= 188.ammp:     pathcc -O3 -OPT:alias=disjoint:unroll_times_max=8:Ofast:ro=3
notes301=               -fno-math-errno -TENV:X=4 +FDO

189.lucas=peak=default=default:
FOPTIMIZE = -O3 -OPT:ro=3:fast_nint=off:unroll_size=256 -WOPT:mem_opnds=on
notes310= 189.lucas:    pathf95 -O3 -OPT:ro=3:fast_nint=off:unroll_size=256
notes311=               -WOPT:mem_opnds=on +FDO

191.fma3d=peak=default=default:
basepeak=1
notes320= 191.fma3d:    pgf90  basepeak=1

200.sixtrack=peak=default=default:
F77OPTIMIZE = -O3 -OPT:Ofast:Olimit=6000:early_intrinsics=on -fno-math-errno -CG:load_exe=1
notes330= 200.sixtrack: pathf95 -O3 -OPT:Ofast:Olimit=6000:early_intrinsics=on
notes331=               -fno-math-errno -CG:load_exe=1 +FDO

301.apsi=peak=default=default:
F77OPTIMIZE = -Ofast -CG:load_exe=0 -LNO:prefetch=0:simd=2
feedback=0
notes340= 301.apsi:     pathf95 -Ofast -CG:load_exe=0 -LNO:prefetch=0:simd=2