#!/bin/bash - false=0; true=1 # Usage: arch [ hostname ] returns hostname, cores, startcore # # Define machine architecture based on starting socket, CPUs (cores) per socket, number of # sockets, has hyperthreading. start=0 arch() { hostname=${1:-`hostname`} # return value hashyper=${true} # assume machine has hyperthreads if [ "${hostname}" = "plg2" ] ; then startsocket=${start} cps=16 # coresPerSocket sockets=2 hashyper=${false} # has no hyperthreads elif [ "${hostname}" = "nasus" ] ; then startsocket=${start} cps=64 # coresPerSocket sockets=2 elif [ "${hostname}" = "pyke" ] ; then startsocket=${start} cps=24 # coresPerSocket sockets=2 elif [ "${hostname}" = "jax" ] ; then startsocket=${start} cps=24 # coresPerSocket sockets=4 else echo "unsupported host" ${hostname} exit 1 fi cores=$(( ${cps} * ${sockets} )) startcore=$(( ${startsocket} * ${cps} )) } # Usage: affinity (global cps, sockets, startsocket, hashyper, cores, startcore, wrap) # returns taskset argument # # This routine assumes hyperthreading has only 2 hyperthreads per core. # # If hyperthread scanning is used: processor units are assigned across the low-number hyperthreads # of the socket's cores. When the low-number hyperthreads are filled, the high-number hyperhtreads # are assigned across the socket's cores. Then the next socket is assigned. # # If hyperthread wrapping is used: processor units are assigned in low/high-number pairs of # hyperthreads across the socket's cores. Then the next socket is assigned. wrap=${false} # set to control hyperthread assignment across socket cores affinity() { if [ ${wrap} -eq ${true} -a ${hashyper} -eq ${false} ] ; then echo "architecture does not support hyperthreading for wrapping" exit 1 fi taskset="" # return value set -- $(( ${1} - 1 )) # decrement $1 if [ ${1} -eq 0 ] ; then taskset="${startcore}-${startcore}"; return; fi if [ ${1} -ge $(( ${cps} * ( ${sockets} - ${startsocket} ) * ( ${hashyper} + 1 ) )) ] ; then # error echo "not enough cores $(( ${cores} * ${sockets} )) for $(( ${1} + 1 )) starting at ${startcore}" exit 1 fi if [ ${hashyper} -eq ${false} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi # no hyperthreads start2=$(( ${startcore} + ${cores} )) if [ ${wrap} -eq ${true} ] ; then # hyperthread wrapping end1=$(( ${1} / 2 + ${startcore} )) end2=$(( ${end1} + ${cores} )) if [ $(( ${1} % 2 )) -eq 0 ] ; then end2=$(( ${end2} - 1 )) fi taskset="${startcore}-${end1},${start2}-${end2}" else # hyperthread scanning if [ ${1} -lt ${cps} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi filled=$(( ${1} / ( ${cps} * 2 ) * ${cps} )) modulus=$(( ${1} % ( ${cps} * 2 ) )) # leftover cores added to saturated sockets if [ ${modulus} -gt ${cps} ] ; then taskset="${startcore}-$(( ${startcore} + ${filled} + ${cps} - 1 )),${start2}-$(( ${start2} + ${filled} + ${modulus} % ${cps} ))" else taskset="${startcore}-$(( ${startcore} + ${filled} + ${modulus} )),${start2}-$(( ${start2} + ${filled} - 1 ))" fi fi } numtimes=5 # numtimes=1 # num_threads='2 4 8 16 24 32' side_chan_threads='6 12 18 24 30' # must be mults of 6 num_threads='2' # side_chan_threads='6' chan_size='10' future_time='10' future_flags=('-DOR' '-DAND3' '-DANDOR' '-DORAND') future_names=('OR' 'AND' 'ANDOR' 'ORAND') # toggle benchmarks spin=${true} contend=${true} sidechan=${true} future=${true} order=${true} spin=${false} contend=${false} sidechan=${false} future=${false} # order=${false} runCFA=${true} runGO=${true} # runUCPP=${true} # runCFA=${false} # runGO=${false} runUCPP=${false} cfa=~/cfa-cc/driver/cfa # Helpers to minimize code duplication # repeats a command ${numtimes} preprint='' repeat_command() { t=1 while [ ${t} -le ${numtimes} ] ; do echo -n -e ${preprint} "${@}" t=`expr ${t} + 1` done } # prints the leading info for a given run of a variant print_header() { echo ${1}':' echo -e "cores\tthroughput (entries)" } # runs the current benchmark with provided args # only works for standard-run benchmarks (not Akka) # must split into pre and post args to be able to supply val of p pre_args='' post_args='' single_run() { affinity ${1} preprint="${1}\t" repeat_command taskset -c ${taskset} ./a.${hostname} ${pre_args} ${1} ${post_args} } # runs the current bench for all processor vals # works for standard benchs that dont need to set a config file (not Akka or CAF) run_bench() { for p in ${num_threads} ; do single_run ${p} done } run_side_chan() { i=1 for p in ${side_chan_threads} ; do affinity ${p} preprint="${p}\t" repeat_command taskset -c ${taskset} ./a.${hostname} ${pre_args} ${i} ${post_args} i=`expr ${i} + 1` done } run_order() { affinity 4 preprint="4\t" repeat_command taskset -c ${taskset} ./a.${hostname} } run_future() { affinity 2 preprint="2\t" repeat_command taskset -c ${taskset} ./a.${hostname} ${post_args} } arch # get hostname # set up leading info for python script echo $numtimes echo $num_threads echo $side_chan_threads if [ ${runCFA} -eq ${true} ]; then echo -n 'CFA ' fi if [ ${runGO} -eq ${true} ]; then echo -n 'Go ' fi echo "" # done printing header info for output # cfa flags cfa_flags='-quiet -O3 -nodebug -DNDEBUG' # UCPP flags UCPPflags="-quiet -g -Wall -Wextra -O3 -nodebug -DNDEBUG -multi" UCPP=~/ucpp/u++-7.0.0/bin/u++ # run the benchmarks run_contend() { post_args=${1} if [ ${runCFA} -eq ${true} ] ; then cd cfa # CFA RUN print_header 'CFA' ${cfa} ${cfa_flags} '-DNUM_CHANS='${3} ${2}.cfa -o a.${hostname} > /dev/null 2>&1 run_bench rm a.${hostname} cd - > /dev/null fi # done CFA if [ ${runGO} -eq ${true} ] ; then cd go/${2}${3} # Go RUN print_header 'Go' go build -o a.${hostname} > /dev/null 2>&1 run_bench rm a.${hostname} cd - > /dev/null fi # done Go } # /usr/bin/time -f "%Uu %Ss %Er %Mkb" if [ ${contend} -eq ${true} ] ; then echo "contend2: " run_contend ${chan_size} 'contend' '2' echo "contend4: " run_contend ${chan_size} 'contend' '4' echo "contend8: " run_contend ${chan_size} 'contend' '8' fi if [ ${spin} -eq ${true} ] ; then echo "spin2: " run_contend ${chan_size} 'spin' '2' echo "spin4: " run_contend ${chan_size} 'spin' '4' echo "spin8: " run_contend ${chan_size} 'spin' '8' fi if [ ${sidechan} -eq ${true} ] ; then echo "sidechan: " post_args=${chan_size} if [ ${runCFA} -eq ${true} ] ; then cd cfa # CFA RUN print_header 'CFA' ${cfa} ${cfa_flags} sidechan.cfa -o a.${hostname} > /dev/null 2>&1 run_side_chan rm a.${hostname} cd - > /dev/null fi # done CFA if [ ${runGO} -eq ${true} ] ; then cd go/sidechan print_header 'Go' go build -o a.${hostname} > /dev/null 2>&1 run_side_chan rm a.${hostname} cd - > /dev/null fi # done Go fi if [ ${future} -eq ${true} ] ; then post_args=${future_time} for i in ${!future_flags[@]}; do echo 'future '${future_names[$i]}':' if [ ${runCFA} -eq ${true} ] ; then cd cfa # CFA RUN print_header 'CFA' ${cfa} ${cfa_flags} ${future_flags[$i]} future.cfa -o a.${hostname} > /dev/null 2>&1 run_future rm a.${hostname} cd - > /dev/null fi # done CFA if [ ${runUCPP} -eq ${true} ] ; then cd ucpp print_header 'uC++' ${UCPP} ${UCPPflags} ${future_flags[$i]} future.cc -o a.${hostname} > /dev/null 2>&1 run_future rm a.${hostname} cd - > /dev/null fi # done Go done fi if [ ${order} -eq ${true} ] ; then echo "order: " post_args=${chan_size} if [ ${runCFA} -eq ${true} ] ; then cd cfa # CFA RUN print_header 'CFA' ${cfa} ${cfa_flags} order.cfa -o a.${hostname} > /dev/null 2>&1 run_order rm a.${hostname} cd - > /dev/null fi # done CFA if [ ${runGO} -eq ${true} ] ; then cd go/order print_header 'Go' go build -o a.${hostname} > /dev/null 2>&1 run_order rm a.${hostname} cd - > /dev/null fi # done Go fi