#!/usr/bin/env sh # Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. # This program is free software; you can redistribute it and/or modify it # under the terms of the license (GNU LGPL) which comes with this package. @CMAKE_SMPI_COMMAND@ SIMGRID_VERSION="@SIMGRID_VERSION_STRING@" SIMGRID_GITHASH="@SIMGRID_GITHASH@" DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps" DEFAULT_LOOPBACK_LATENCY="0.000004s" DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))Bps" DEFAULT_NETWORK_LATENCY="0.000005s" DEFAULT_NUMPROCS="4" DEFAULT_SPEED="100flops" LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}" LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}" NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}" NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}" SPEED="${DEFAULT_SPEED}" PRIVATIZE="--cfg=smpi/privatization:${SMPI_PRIVATIZATION:-@HAVE_PRIVATIZATION@}" SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI" #usage to print the way this script should be called usage () { cat < -hostfile program [program-options] $0 [OPTIONS] -platform -hostfile -replay [program] [program-options] Options: -keep-temps # don't remove the generated files after execution -wrapper # use command to run the program (e.g. "valgrind" or "gdb --args") -gdb # run within GDB (-wrapper "gdb --args" -keep-temps) -lldb # run within LLDB (-wrapper "lldb --" -keep-temps) -vgdb # run within Valgrind+GDB (-wrapper "valgrind --vgdb=yes --vgdb-error=0" -keep-temps) -map # display the machine on which each process rank is mapped -np # use that amount of processes from the hostfile. # By default, all processes of the hostfile are used. -no-privatize # Disable the globals privatization, that is activated by default -trace-ti # activate time independant tracing (for replay, default in smpi_simgrid.txt) -trace # activate tracing (Paje, default in smpi_simgrid.trace) -trace-comment # put a comment on the top of the trace file -trace-comment-file # put file contents on the top of the trace file as comment -trace-grouped # group MPI processes by location -trace-resource # trace resource utilization -trace-file # name of the tracefile (simgrid_smpi.trace) -replay # replays a trace instead of actually executing an application -version # Displays the SimGrid version (human readable) -git-version # Displays the git hash of SimGrid or (deprecated usage): $0 [-keep-temps] [-np ] [-bandwidth ] [-latency ] program [program-options] EOF } #check if we have at least one parameter if [ $# -eq 0 ] then usage exit fi WRAPPER="" HOSTFILE="" HOSTFILETMP=0 MAPOPT=0 REPLAY=0 QUIET="" unset pid trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS" die () { printf '[%s] ** error: %s. Aborting.\n' "$(basename $0)" "$*" >&2 exit 1 } smpirun_cleanup() { if [ -z "${KEEP}" ] ; then if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then rm -f ${PLATFORMTMP} PLATFORMTMP="" fi if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then rm -f ${HOSTFILE} HOSTFILE="" fi if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then rm -f ${UNROLLEDHOSTFILE} UNROLLEDHOSTFILE="" fi if [ -n ${APPLICATIONTMP} ]; then rm -f ${APPLICATIONTMP} APPLICATIONTMP="" fi fi } smpirun_trap() { local sig sig="$1" # Cleanup and kill the child process: smpirun_cleanup if ! [ -z "$pid" ]; then kill -TERM $pid fi unset pid # Raise the same signal again (remove the traps first): trap - $trapped_signals kill -$sig $$ # This should never happen: kill -ABRT $$ kill -TERM $$ } for s in $trapped_signals; do trap "smpirun_trap $s" $s done while true; do case "$1" in "-np" | "-n") NUMPROCS="$2" shift 2 ;; "-bandwidth") NETWORK_BANDWIDTH="$2" shift 2 ;; "-latency") NETWORK_LATENCY="$2" shift 2 ;; "-platform") PLATFORM="$2" if [ ! -f "${PLATFORM}" ]; then die "the file '${PLATFORM}' does not exist" fi shift 2 ;; "-hostfile" | "-machinefile") HOSTFILE="$2" if [ ! -f "${HOSTFILE}" ]; then die "the file '${HOSTFILE}' does not exist" fi shift 2 ;; "-replay") APP_TRACES="$2" shift 2 ;; "-no-privatize") PRIVATIZE="--cfg=smpi/privatization:no" shift 1 ;; "-map") MAPOPT=1 shift 1 ;; "-trace") TRACE_ACTIVE="true" shift 1 ;; "-trace-ti") TRACE_ACTIVE="true" TRACE_TI_ACTIVE="true" shift 1 ;; "-trace-comment") TRACE_COMMENT="$2" shift 2 ;; "-trace-comment-file") TRACE_COMMENT_FILE="$2" shift 2 ;; "-trace-file") TRACE_FILENAME="$2" shift 2 ;; "-trace-grouped") TRACE_GROUPED="true" shift 1 ;; "-trace-resource") TRACE_RESOURCE="true" shift 1 ;; "-keep-temps") KEEP="true" SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" shift 1 ;; "-quiet") QUIET="true" shift 1 ;; "-wrapper") WRAPPER="$2" shift 2 ;; "-gdb") WRAPPER="gdb --args" KEEP="true" SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" shift 1 ;; "-vgdb") WRAPPER="valgrind --vgdb=yes --vgdb-error=0" KEEP="true" SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" shift 1 ;; "-lldb") WRAPPER="lldb --" KEEP="true" SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" shift 1 ;; "-help" | "--help" | "-h") usage exit 0 ;; "-version" | "--version" | "-v") printf '%b\n' "$SIMGRID_VERSION" exit 0 ;; "-git-version" | "--git-version") printf '%b\n' "$SIMGRID_GITHASH" exit 0 ;; "--cfg="*|"--log="*) for OPT in ${1#*=} do SIMOPTS="$SIMOPTS ${1%%=*}=$OPT" done shift 1 ;; "-foreground") # Nothing to do, compatibility. shift 1 ;; *) break ;; esac done if [ -n "${APP_TRACES}" ] ; then if [ $# -eq 0 ] ; then EXEC="@SMPIREPLAYMAIN@" else EXEC="$1" shift fi else # check if we still have at least one parameter beyond options if [ $# -eq 0 ] then echo "Error: no program to execute!" usage exit fi EXEC="$1" shift fi # steel --cfg and --logs options while [ $# -gt 0 ]; do case "$1" in "--cfg="*|"--log="*) for OPT in ${1#*=} do SIMOPTS="$SIMOPTS ${1%%=*}=$OPT" done shift 1 ;; *) PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1" shift ;; esac done if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then echo "No hostfile nor platform specified." usage exit 1 fi if [ -z "${HOSTFILE}" ] ; then HOSTFILETMP=1 HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)" @PYTHON_EXECUTABLE@ -c ' import xml.etree.ElementTree as ET import sys import re tree = ET.parse(sys.stdin) for elem in tree.findall(".//host"): print(elem.attrib["id"]) for elem in tree.findall(".//cluster"): prefix = elem.attrib["prefix"] radical = elem.attrib["radical"] suffix = elem.attrib["suffix"] for r in radical.split(","): m = re.match("^([^-]*?)-([^-]*)$", r) if m: for i in range(int(m.group(1)), int(m.group(2))): print(prefix + str(i) + suffix) else: print(prefix + r + suffix) ' < ${PLATFORM} > ${HOSTFILE} fi UNROLLEDHOSTFILETMP=0 # parse if our lines are terminated by :num_process if grep -q ':' $HOSTFILE ; then UNROLLEDHOSTFILETMP=1 UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)" @PYTHON_EXECUTABLE@ -c ' import sys import re for line in sys.stdin: m = re.match("(.*):(.*)", line) if m: for i in range(0, int(m.group(2))): print(m.group(1)) else: print(line.strip()) ' < ${HOSTFILE} > ${UNROLLEDHOSTFILE} if [ ${HOSTFILETMP} = 1 ] ; then rm ${HOSTFILE} HOSTFILETMP=0 fi HOSTFILE=$UNROLLEDHOSTFILE fi # Don't use wc -l to compute it to avoid issues with trailing \n at EOF hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE) if [ ${hostfile_procs} = 0 ] ; then die "the hostfile '${HOSTFILE}' is empty" fi if [ -z "${NUMPROCS}" ] ; then # Use the amount of processes in the hostfile as default value for the -np parameter NUMPROCS=$hostfile_procs fi if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2 fi ##-------------------------------- DEFAULT or SPECIFIED PLATFORM -------------------------------------- if [ -z "${PLATFORM}" ]; then PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)" cat > ${PLATFORMTMP} < PLATFORMHEAD i=${NUMPROCS} while [ $i -gt 0 ]; do echo " " >> ${PLATFORMTMP} echo " " >> ${PLATFORMTMP} echo " " >> ${PLATFORMTMP} i=$((i - 1)) done i=${NUMPROCS} while [ $i -gt 0 ]; do j=${NUMPROCS} while [ $j -gt 0 ]; do if [ $i -eq $j ]; then echo " " >> ${PLATFORMTMP} else echo " " >> ${PLATFORMTMP} fi j=$((j - 1)) done i=$((i - 1)) done cat >> ${PLATFORMTMP} < PLATFORMFOOT else PLATFORMTMP=${PLATFORM} fi ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM -------------------------------------- ##-------------------------------- DEFAULT APPLICATION -------------------------------------- APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)" #APPLICATIONTMP="app.xml" cat > ${APPLICATIONTMP} < APPLICATIONHEAD ##---- cache hostnames of hostfile--------------- if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then hostnames=$(cat ${HOSTFILE} | tr '\n\r' ' ') fi if [ -n "${APP_TRACES}" ]; then if [ -f "${APP_TRACES}" ]; then hosttraces=$(cat ${APP_TRACES} | tr '\n\r' ' ' ) NUMTRACES=$(cat ${APP_TRACES} | wc -l) REPLAY=1 else printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2 exit 1 fi fi ##---------------------------------------------------------- ## generate application.xml with hostnames from hostfile: ## the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if ## hostfile has less than i lines. ##---------------------------------------------------------- HAVE_SEQ="$(which seq 2>/dev/null)" if [ -n "${HAVE_SEQ}" ]; then SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1))) else cnt=0 while [ $cnt -lt ${NUMPROCS} ] ; do SEQ="$SEQ $cnt" cnt=$((cnt + 1)); done fi set -- $hostnames ##---- generate tags------------------------------ #prepare arguments at once for ARG in $PROC_ARGS; do XML_ARGS="${XML_ARGS}"" " done for i in ${SEQ} do j=$(( i % hostfile_procs + 1 )) host=$(eval "echo \${$j}") ##---- optional display of ranks to actor mapping if [ ${MAPOPT} = 1 ]; then echo "[rank $i] -> $host" fi echo " " >> ${APPLICATIONTMP} if [ ${REPLAY} = 1 ]; then echo " " >> ${APPLICATIONTMP} if [ ${NUMTRACES} -gt 1 ]; then echo " " >> ${APPLICATIONTMP} else echo " " >> ${APPLICATIONTMP} fi else echo ${XML_ARGS} >> ${APPLICATIONTMP} fi echo " " >> ${APPLICATIONTMP} done cat >> ${APPLICATIONTMP} < APPLICATIONFOOT ##-------------------------------- end DEFAULT APPLICATION -------------------------------------- ##---------------------- SMPI TRACING OPTIONS --------------------------------- if [ -n "${TRACE_ACTIVE}" ]; then #define trace filename if [ -n "${TRACE_TI_ACTIVE}" ]; then if [ -z "${TRACE_FILENAME}" ]; then TRACE_FILENAME="smpi_simgrid.txt" fi TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes" else if [ -z "${TRACE_FILENAME}" ]; then TRACE_FILENAME="smpi_simgrid.trace" fi TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes" fi if [ -n "${TRACE_COMMENT}" ]; then TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}" fi if [ -n "${TRACE_COMMENT_FILE}" ]; then TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}" fi if [ -n "${TRACE_GROUPED}" ]; then TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes" fi if [ -n "${TRACE_RESOURCE}" ]; then TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes" fi fi ##---------------------- end SMPI TRACING OPTIONS --------------------------------- # Do not remove, this variable may be used by user code (e.g. StarPU) export SMPI_GLOBAL_SIZE=${NUMPROCS} if [ -n "${KEEP}" ] ; then echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} if [ ${HOSTFILETMP} = 1 ] ; then echo "Generated hostfile ${HOSTFILE} kept." fi if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept." fi fi # Execute the process # # The shell still need to be alive for the duration in order to do some cleanup after the process. # # We are going through great lengths in order to both keep stdin and be able to handle signals: # # * The job is launched in the background in order to be able to handle signals. # # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command # is launched in the background: this can be overriden in bash but not in standard bourne shell. exec 3<&0 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- & pid=$! exec 3>&- wait $pid status=$? # With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit # status of 128. Try again. while test $status -eq 128 && kill -0 $pid 2>/dev/null; do wait $pid status=$? done pid="" # Keep temporary files on failures to help debugging # if [ ${status} -ne 0 ] ; then if [ -z "${KEEP}" -a -z "${QUIET}" ]; then echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} if [ ${HOSTFILETMP} = 1 ] ; then echo "Generated hostfile ${HOSTFILE} kept." fi if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept." fi KEEP=true fi echo "Execution failed with code ${status}." fi smpirun_cleanup exit $status