# Invocation command line:
# /work/cpu2006/bin/runspec -c fsc-mixpp-20080516-rate -T all --input ref -n 3 --rate 8 --define binding -o all fp
# output_root was not used for this run
############################################################################
#  CPU2006 configuration file for 64-bit Linux on AMD Quad-core.
#  Intended for rate only.
#
#  May 2008
#
#      Compiler name/version:       PGI 7.2
#                                   Pathscale 3.2
#      Operating system version:    64-bit Linux
#      Hardware:                    AMD Opteron (Quad-core)
#      FP Base Pointer Size:        64-bit only
#      FP Peak Pointer Size:        32/64-bit only
#      INT Base Pointer Size:       32/64-bit
#      INT Peak Pointer Size:       32/64-bit
#      Auto Parallization:          Not Used
#

#####################################################################
# Macro section
####################################################################
# Modify this section to use the appropriate architecture flags
%define pgi_tp64 -tp barcelona-64
%define pgi_tp32 -tp barcelona

# Change this to 896 for systems with 4GB of memory per copy
%define pgi_hp        150

# Adjust the build jobs to the number of concurrent build processses
%define build_jobs 4

#####################################################################
# Header Section 
#####################################################################
ext           = pgi_path_barcelona_rate_20080516
ignore_errors = no
tune          = base,peak
output_format = asc,pdf,raw,flags,cfg,html
flagsurl000   = http://www.spec.org/cpu2006/flags/fsc-mix-pgi-path.xml
size          = test,train,ref
check_md5     = yes
reportable    = yes
mean_anyway   = yes
env_vars      = no
# teeout        = yes
teerunout     = yes
#output_root   = /local/home/${username}/CPU2006
#allow_extension_override = yes
makeflags     = -j %{build_jobs}

####################################################################
# Include file containing the SUT hardware inforamtion
# as well as the submit command, tester information and notes
####################################################################
#include: SUT.inc
#  ----- Begin inclusion of 'SUT.inc'
############################################################################
# File to be included into config file for SPEC CPU2000

# This include file specifies the "System Under Test" (Hardware,
# OS, BIOS, etc.)
# Any information about the binaries is elsewhere.

# To avoid conflicts in the numbering of the notes,
# please use only 900 and higher numbers.

%ifdef %{binding}
use_submit_for_speed=1

# submit= MYMASK=\$((1<<\$SPECUSERNUM)) ; /usr/bin/taskset \$MYMASK $command
# submit=MYMASK=`printf '0x%x' \$((1<<$SPECUSERNUM))` ; /usr/bin/taskset \$MYMASK $command

# submit mit numactl
# submit=VALCORE=`printf '%d' \$SPECCOPYNUM` ; VALMEM=`printf '%d' \$((\$SPECCOPYNUM/2))` ; numactl --membind \$VALMEM taskset -c \$VALCORE $command

bind0 = numactl -m 0 --physcpubind=0
bind1 = numactl -m 0 --physcpubind=1
bind2 = numactl -m 0 --physcpubind=2
bind3 = numactl -m 0 --physcpubind=3
bind4 = numactl -m 1 --physcpubind=4
bind5 = numactl -m 1 --physcpubind=5
bind6 = numactl -m 1 --physcpubind=6
bind7 = numactl -m 1 --physcpubind=7

submit = echo "$command" >run.sh ; $BIND bash run.sh

notes_005=
notes_010=The command numactl has been used to bind processes to CPUs
# notes911='taskset $MYMASK $command' has been used to bind processes to CPUs
# notes912=The Linux command 'taskset $MYMASK $command' executes the specified command;
# notes913=only the processors specified in $MYMASK are allowed to execute the application.
# notes914=See the Linux documentation for detailed description of taskset.
%endif

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

default=default=default=default:
notes_os_000= powersave -f is applied to set CPU to maximum frequency prior to run
notes_os_005= stacksize is set to unlimited prior to run
notes_os_010= ulimit -l 2457600
notes_os_015= PGI_HUGE_PAGES set to 150
notes_os_020= (Total number of huge pages available is 1200)
notes_000=
# notes901=BIOS default settings with
# notes902=Node memory interleave disabled, SRAT enabled

################################################################
# About Fujitsu Siemens Computers
################################################################

default=default=default=default:
notes_015=
notes_020=For information about Fujitsu Siemens Computers please see:
notes_025=http://www.fujitsu-siemens.com

# Blank line at the end
notes_030=

################################################################
# Description Hardware and Software
################################################################

default=default=default=default:
hw_vendor=Fujitsu Siemens Computers
hw_model=CELSIUS V840, AMD Opteron 2354 (2.2 GHz)
hw_cpu_name=AMD Opteron 2354
hw_cpu_char=
hw_cpu_mhz=2200
hw_fpu=Integrated

hw_nchips=2
hw_ncores=8
hw_ncoresperchip=4
hw_nthreadspercore=1

hw_ncpuorder=1,2 chips
hw_pcache=64 KB I + 64 KB D on chip per core
hw_scache=512 KB I+D on chip per core
hw_tcache=2 MB I+D on chip per chip
hw_ocache=None
hw_memory=16 GB (8x2GB PC2-5300P, CL5, dual rank ECC)
hw_disk=1 x 400 GB SATA II, 7200 rpm
hw_other=None

sw_file=ext3
sw_state=Multi-User SuSE Run Level 3

test_sponsor=Fujitsu Siemens Computers
license_num=22
tester=Fujitsu Siemens Computers
test_date=May-2008
hw_avail=May-2008
prepared_by=Fujitsu Siemens Computers
# ---- End inclusion of '/work/cpu2006/config/SUT.inc'

############################ Software Info   ############################
default=default=default=default:
sw_os000     = SUSE Linux Enterprise Server 10 (x86_64) SP1,
sw_os001     = Kernel 2.6.16.46-0.12-smp
sw_other000  = binutils 2.18.50

fp=default=default=default:
sw_peak_ptrsize  = 32/64-bit
sw_base_ptrsize  = 64-bit
sw_auto_parallel = No
sw_avail         = May-2008
sw_compiler000   = PGI Workstation Complete Version 7.2-1
sw_compiler001   = PathScale Compiler Suite, Release 3.2 Beta

int=default=default=default:
sw_peak_ptrsize  = 32/64-bit
sw_base_ptrsize  = 32/64-bit
sw_auto_parallel = No
sw_other001      = Microquill SmartHeap 8.1 32-bit Library for Linux
sw_avail         = May-2008
sw_compiler001   = PGI Workstation Complete Version 7.2-1
sw_compiler002   = PathScale Compiler Suite, Release 3.2 Beta

default=default=default=default:
#####################################################################
# Compiler selection

# Default: PGI compilers

CC           = pgcc 
CXX          = pgcpp 
FC           = pgf95

# For some programs in peak: Pathscale compilers

CCPATH  = pathcc -march=barcelona
CXXPATH = pathCC -march=barcelona
FCPATH  = pathf95 -march=barcelona

SMARTHEAP_DIR = /opt/SmartHeap_8.1/lib

#####################################################################
# Optimization
#####################################################################
default=default=default=default:
EXTRA_LDFLAGS   = -Bstatic_pgi

#####################################################################
# BASE FLAGS
#####################################################################

fp=base=default=default:
COPTIMIZE    = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
FOPTIMIZE    = -fastsse -Mfprelaxed -Msmartalloc=huge:%{pgi_hp} -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
CXXOPTIMIZE  = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed --zc_eh -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 

int=base=default=default:
COPTIMIZE    = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
CXXOPTIMIZE  = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed --zc_eh -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp32} 

#####################################################################
# PEAK FLAGS
#####################################################################

400.perlbench=peak=default:
CC=$(CCPATH)
EXTRA_LDFLAGS =
COPTIMIZE=-Ofast -IPA:plimit=20000 -LNO:opt=0 -WOPT:if_conv=0 -CG:local_sched_alg=1
PASS1_CFLAGS  = -fb_create fbdata 
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata
feedback      = 1
basepeak      = 0

401.bzip2=peak=default=default:
COPTIMIZE     = -fastsse -O4 -Msmartalloc=huge:%{pgi_hp} -Mprefetch=t0 -Mnounroll  %{pgi_tp64} 
PASS1_CFLAGS  = -Mpfi=indirect
PASS1_LDFLAGS = -Mpfi=indirect
PASS2_CFLAGS  = -Mpfo=indirect 
PASS2_LDFLAGS = -Mpfo=indirect 
feedback      = 1
basepeak      = 0

403.gcc=peak=default:
CC=$(CCPATH)
EXTRA_LDFLAGS =
PORTABILITY =
COPTIMIZE=-O3 -OPT:Ofast -m32
PASS1_CFLAGS  = -fb_create fbdata 
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata
feedback      = 1
basepeak      = 0

429.mcf=peak=default=default:
COPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mipa=jobs:%{build_jobs},fast,inline:1  %{pgi_tp32} 
feedback      = 0
basepeak      = 0

445.gobmk=peak=default:
CC=$(CCPATH)
EXTRA_LDFLAGS =
COPTIMIZE=-O3 -OPT:alias=restrict -LNO:prefetch=1:ignore_feedback=off -CG:p2align=on
PASS1_CFLAGS  = -fb_create fbdata 
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata
feedback      = 1
basepeak      = 0

456.hmmer=peak=default=default:
COPTIMIZE     = -fastsse -Mvect=partial -Munroll=n:8 -Msmartalloc=huge:%{pgi_hp} -Msafeptr -Mprefetch=t0 -Mfprelaxed    -Mipa=jobs:%{build_jobs},const,ptr,arg,inline   %{pgi_tp64} 
feedback      = 0
basepeak      = 0

458.sjeng=peak=default=default:
COPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed     %{pgi_tp64} 
PASS1_CFLAGS  = -Mpfi 
PASS1_LDFLAGS = -Mpfi 
PASS2_CFLAGS  = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline:1,noarg  
PASS2_LDFLAGS = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline:1,noarg  
feedback      = 1
basepeak      = 0

462.libquantum=peak=default=default:
COPTIMIZE     = -fastsse -Munroll=m:8 -Msmartalloc=huge:%{pgi_hp} -Mprefetch=distance:4 -Mfprelaxed -Mipa=jobs:%{build_jobs},fast,inline,noarg  %{pgi_tp64} 
feedback      = 0
basepeak      = 0

464.h264ref=peak=default:
CC=$(CCPATH)
EXTRA_LDFLAGS =
COPTIMIZE=-O3 -IPA:plimit=20000 -OPT:alias=disjoint -LNO:prefetch=0 -CG:ptr_load_use=0:push_pop_int_saved_regs=off
PASS1_CFLAGS  = -fb_create fbdata 
PASS1_LDFLAGS = -fb_create fbdata
PASS2_CFLAGS  = -fb_opt fbdata
PASS2_LDFLAGS = -fb_opt fbdata
feedback      = 1
basepeak      = 0

471.omnetpp=peak=default=default:
#CXXOPTIMIZE   = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed --zc_eh -Mipa=jobs:%{build_jobs},fast,inline %{pgi_tp32} 
feedback      = 0
basepeak      = 1

473.astar=peak=default=default:
CXXOPTIMIZE   = -fastsse -O4 -Msmartalloc=huge:%{pgi_hp} -Msafeptr=global -Mfprelaxed --zc_eh  %{pgi_tp32} 
PASS1_CXXFLAGS= -Mpfi 
PASS1_LDFLAGS = -Mpfi 
PASS2_CXXFLAGS= -Mpfo -Mipa=jobs:%{build_jobs},fast,inline:6  
PASS2_LDFLAGS = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline:6  
feedback      = 1
basepeak      = 0

483.xalancbmk=peak=default:
CXX=$(CXXPATH)
EXTRA_LDFLAGS =
CXXOPTIMIZE=-Ofast -OPT:unroll_times_max=8 -CG:push_pop_int_saved_regs=off:ptr_load_use=0 -m32
EXTRA_CXXLIBS= -L$(SMARTHEAP_DIR) -lsmartheap
feedback      = 0
basepeak      = 0

410.bwaves=peak=default=default:
FOPTIMIZE     = -fastsse -Msmartalloc -Mprefetch=distance:12,nta -Mpre -Mfprelaxed  %{pgi_tp64} 
PASS1_FFLAGS  = -Mpfi 
PASS1_LDFLAGS = -Mpfi 
PASS2_FFLAGS  = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline 
PASS2_LDFLAGS = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline 
feedback      = 1
basepeak      = 0

416.gamess=peak=default:
FC = $(FCPATH)
EXTRA_LDFLAGS =
FOPTIMIZE=-O2 -OPT:Ofast:ro=3:unroll_size=256
PASS1_FFLAGS   = -fb_create fbdata
PASS2_FFLAGS   = -fb_opt fbdata
PASS1_LDFLAGS  = -fb_create fbdata
PASS2_LDFLAGS  = -fb_opt fbdata
feedback      = 1
basepeak      = 0

433.milc=peak=default=default:
COPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Msafeptr -Mfprelaxed -Mipa=jobs:%{build_jobs},inline,arg,const,ptr,shape   %{pgi_tp64} 
feedback      = 0
basepeak      = 0

434.zeusmp=peak=default=default:
#FOPTIMIZE     = -fastsse -Mfprelaxed -Msmartalloc -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
feedback      = 0
basepeak      = 1

435.gromacs=peak=default=default:
FOPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed -Mfpapprox=rsqrt -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
COPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mfprelaxed -Mfpapprox=rsqrt -Mipa=jobs:%{build_jobs},fast,inline  %{pgi_tp64} 
feedback      = 0
basepeak      = 0

436.cactusADM=peak=default:
FC = $(FCPATH)
CC = $(CCPATH)
EXTRA_LDFLAGS =
FOPTIMIZE=-Ofast -LNO:blocking=off
COPTIMIZE=$(FOPTIMIZE)
PASS1_CFLAGS   = -fb_create fbdata
PASS2_CFLAGS   = -fb_opt fbdata
PASS1_FFLAGS   = -fb_create fbdata
PASS2_FFLAGS   = -fb_opt fbdata
PASS1_LDFLAGS  = -fb_create fbdata
PASS2_LDFLAGS  = -fb_opt fbdata
feedback      = 1
basepeak      = 0

437.leslie3d=peak=default=default:
FOPTIMIZE     = -fastsse -Mvect=fuse -Msmartalloc=huge:%{pgi_hp} -Mprefetch=distance:8,t0 -Mfprelaxed  %{pgi_tp64} 
PASS1_FFLAGS  = -Mpfi=indirect
PASS1_LDFLAGS = -Mpfi=indirect 
PASS2_FFLAGS  = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline 
PASS2_LDFLAGS = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline 
feedback      = 1
basepeak      = 0

444.namd=peak=default=default:
CXXOPTIMIZE   = -fastsse -Munroll=n:4,m:8 -Msmartalloc=huge:%{pgi_hp} -Mnodepchk -Mfprelaxed --zc_eh  %{pgi_tp64} 
PASS1_CXXFLAGS= -Mpfi 
PASS1_LDFLAGS = -Mpfi 
PASS2_CXXFLAGS= -Mpfo -Mipa=jobs:%{build_jobs},fast,inline 
PASS2_LDFLAGS = -Mpfo -Mipa=jobs:%{build_jobs},fast,inline 
feedback      = 1
basepeak      = 0

447.dealII=peak=default:
# Needed to avoid -DSPEC_CPU_LP64
PORTABILITY =
CXX = $(CXXPATH)
EXTRA_LDFLAGS =
CXXOPTIMIZE=-Ofast -static -INLINE:aggressive=on -fno-exceptions -m32
LDCXXFLAGS=-lm
feedback      = 0
basepeak      = 0

450.soplex=peak=default:
# Needed to avoid -DSPEC_CPU_LP64 
PORTABILITY =
CXX = $(CXXPATH)
EXTRA_LDFLAGS =
CXXOPTIMIZE=-O3 -TENV:frame_pointer=off -LNO:prefetch=1 -OPT:malloc_alg=1 -CG:load_exe=0 -m32
PASS1_CXXFLAGS = -fb_create fbdata
PASS2_CXXFLAGS = -fb_opt fbdata
PASS1_LDFLAGS  = -fb_create fbdata
PASS2_LDFLAGS  = -fb_opt fbdata
feedback      = 1
basepeak      = 0

453.povray=peak=default:
CXX = $(CXXPATH)
EXTRA_LDFLAGS =
CXXOPTIMIZE = -Ofast 
PASS1_CXXFLAGS = -fb_create fbdata
PASS2_CXXFLAGS = -fb_opt fbdata
PASS1_LDFLAGS  = -fb_create fbdata
PASS2_LDFLAGS  = -fb_opt fbdata
feedback      = 1
basepeak      = 0

454.calculix=peak=default=default:
FOPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mprefetch=t0 -Mpre -Mfprelaxed  %{pgi_tp64} 
COPTIMIZE     = -fastsse -Msmartalloc=huge:%{pgi_hp} -Mprefetch=t0 -Mpre -Mfprelaxed  %{pgi_tp64} 
PASS1_FFLAGS  = -Mpfi=indirect 
PASS1_CFLAGS  = -Mpfi=indirect 
PASS1_LDFLAGS = -Mpfi=indirect 
PASS2_FFLAGS  = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline 
PASS2_CFLAGS  = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline 
PASS2_LDFLAGS = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline 
feedback      = 1
basepeak      = 0

459.GemsFDTD=peak=default:
FC = $(FCPATH)
EXTRA_LDFLAGS =
FOPTIMIZE=-Ofast -LNO:fission=2:simd=2:prefetch_ahead=1 -CG:load_exe=0
feedback      = 0
basepeak      = 0

465.tonto=peak=default:
FC = $(FCPATH)
EXTRA_LDFLAGS =
FOPTIMIZE=-Ofast -OPT:alias=no_f90_pointer_alias -LNO:blocking=off -CG:load_exe=1 -IPA:plimit=525
feedback      = 0
basepeak      = 0

470.lbm=peak=default:
CC = $(CCPATH)
EXTRA_LDFLAGS =
COPTIMIZE=-Ofast -CG:sse_cse_regs=0 -CG:locs_shallow_depth=1 -m3dnow
feedback      = 0
basepeak      = 0

481.wrf=peak=default=default:
FOPTIMIZE     = -fastsse -Mvect=noaltcode -Msmartalloc -Mprefetch=distance:8 -Mfprelaxed  %{pgi_tp64} 
COPTIMIZE     = -fastsse -Mvect=noaltcode -Msmartalloc -Mprefetch=distance:8 -Mfprelaxed  %{pgi_tp64} 
feedback      = 0
basepeak      = 0

482.sphinx3=peak=default=default:
COPTIMIZE     = -fastsse -Mfprelaxed -Msmartalloc %{pgi_tp64} 
PASS1_FFLAGS  = -Mpfi=indirect
PASS1_LDFLAGS = -Mpfi=indirect 
PASS2_FFLAGS  = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline
PASS2_LDFLAGS = -Mpfo=indirect -Mipa=jobs:%{build_jobs},fast,inline
feedback      = 1
basepeak      = 0

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

fp=default=default=default:
PORTABILITY     = -DSPEC_CPU_LP64

int=default=default=default:
CPORTABILITY    = -DSPEC_CPU_LP64

400.perlbench=default=default=default:
CPORTABILITY    =  -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX_X64 

403.gcc=peak=default=default:
CPORTABILITY    =

429.mcf=peak=default=default:
CPORTABILITY    = 

435.gromacs=default=default=default:
LDPORTABILITY   = -Mnomain

# For use with PGI compiler mix
436.cactusADM=base=default=default:
LDPORTABILITY   = -Mnomain

# For use with Pathscale compiler mix
436.cactusADM=peak=default:
FPORTABILITY=  -fno-second-underscore		   

447.dealII=peak=default=default:
PORTABILITY= 

454.calculix=default=default=default:
LDPORTABILITY   = -Mnomain

462.libquantum=default=default=default:
CPORTABILITY    = -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX

481.wrf=default=default=default:
CPORTABILITY    = -DSPEC_CPU_CASE_FLAG -DSPEC_CPU_LINUX 

483.xalancbmk=default=default=default:
CXXPORTABILITY  = -DSPEC_CPU_LINUX