3 # Copyright (c) 2007-2018. The SimGrid Team. All rights reserved.
5 # This program is free software; you can redistribute it and/or modify it
6 # under the terms of the license (GNU LGPL) which comes with this package.
10 SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
11 SIMGRID_GITHASH="@SIMGRID_GITHASH@"
13 DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps"
14 DEFAULT_LOOPBACK_LATENCY="0.000004s"
15 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))Bps"
16 DEFAULT_NETWORK_LATENCY="0.000005s"
18 DEFAULT_SPEED="100flops"
20 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
21 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
22 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
23 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
24 SPEED="${DEFAULT_SPEED}"
26 PRIVATIZE="--cfg=smpi/privatization:${SMPI_PRIVATIZATION:-@HAVE_PRIVATIZATION@}"
28 SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI"
30 #usage to print the way this script should be called
33 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
35 -keep-temps # don't remove the generated files after execution
36 -wrapper <command> # use command to run the program (e.g. "valgrind" or "gdb --args")
37 -map # display the machine on which each process rank is mapped
38 -np <numprocs> # use that amount of processes from the hostfile.
39 # By default, all processes of the hostfile are used.
40 -no-privatize # Disable the globals privatization, that is activated by default
41 -trace-ti # activate time independant tracing (for replay, default in smpi_simgrid.txt)
42 -trace # activate tracing (Paje, default in smpi_simgrid.trace)
43 -trace-comment <comment> # put a comment on the top of the trace file
44 -trace-comment-file <file> # put file contents on the top of the trace file as comment
45 -trace-grouped # group MPI processes by location
46 -trace-resource # trace resource utilization
47 -trace-file <tracefile> # name of the tracefile (simgrid_smpi.trace)
48 -ext <value> # additional parameter (reserved)
50 -version # Displays the SimGrid version (human readable)
51 -git-version # Displays the git hash of SimGrid
53 or (deprecated usage):
54 $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
59 #check if we have at least one parameter
75 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
78 printf '[%s] ** error: %s. Aborting.\n' "$(basename $0)" "$*" >&2
84 if [ -z "${KEEP}" ] ; then
85 if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
89 if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
93 if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
94 rm -f ${UNROLLEDHOSTFILE}
97 if [ -n ${APPLICATIONTMP} ]; then
98 rm -f ${APPLICATIONTMP}
108 # Cleanup and kill the child process:
110 if ! [ -z "$pid" ]; then
115 # Raise the same signal again (remove the traps first):
116 trap - $trapped_signals
119 # This should never happen:
124 for s in $trapped_signals; do
125 trap "smpirun_trap $s" $s
135 NETWORK_BANDWIDTH="$2"
144 if [ ! -f "${PLATFORM}" ]; then
145 die "the file '${PLATFORM}' does not exist"
149 "-hostfile" | "-machinefile")
151 if [ ! -f "${HOSTFILE}" ]; then
152 die "the file '${HOSTFILE}' does not exist"
161 PRIVATIZE="--cfg=smpi/privatization:no"
174 TRACE_TI_ACTIVE="true"
181 "-trace-comment-file")
182 TRACE_COMMENT_FILE="$2"
194 TRACE_RESOURCE="true"
199 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
206 "-help" | "--help" | "-h")
210 "-version" | "--version" | "-v")
211 printf '%b\n' "$SIMGRID_VERSION"
214 "-git-version" | "--git-version")
215 printf '%b\n' "$SIMGRID_GITHASH"
221 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
226 # Nothing to do, compatibility.
235 # check if we still have at least one parameter beyond options
238 echo "Error: no program to execute!"
246 # steel --cfg and --logs options
247 while [ $# -gt 0 ]; do
252 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
257 PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
263 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
264 echo "No hostfile nor platform specified."
269 if [ -z "${HOSTFILE}" ] ; then
271 HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
272 perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
273 perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
274 my ($pre,$rad,$post)=($1,$2,$3);
275 for my $elm (split(",",$rad)) {
276 if ($elm=~/^([^-]*?)-([^-]*)$/) {
277 for (my $i=$1; $i<=$2;$i++) {
278 print "$pre$i$post\n";
281 print "$pre$elm$post\n";
284 } elsif (/<cluster/) {
285 die ("Unparsable cluster tag. smpirun uses a primitive regular expression to parse cluster tags. Either provide a hostfile yourself or give the attributes prefix, radical and suffix IN THAT ORDER and ON THE SAME LINE as the opening tag (<cluster)");
286 }' ${PLATFORM} >> ${HOSTFILE}
288 UNROLLEDHOSTFILETMP=0
290 #parse if our lines are terminated by :num_process
291 multiple_processes=$(grep -c ":" $HOSTFILE)
292 if [ "${multiple_processes}" -gt 0 ] ; then
293 UNROLLEDHOSTFILETMP=1
294 UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
295 perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/' ${HOSTFILE} > ${UNROLLEDHOSTFILE}
296 if [ ${HOSTFILETMP} = 1 ] ; then
300 HOSTFILE=$UNROLLEDHOSTFILE
303 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
304 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
305 if [ ${hostfile_procs} = 0 ] ; then
306 die "the hostfile '${HOSTFILE}' is empty"
309 if [ -z "${NUMPROCS}" ] ; then
310 # Use the amount of processes in the hostfile as default value for the -np parameter
311 NUMPROCS=$hostfile_procs
314 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
315 echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
318 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
319 if [ -z "${PLATFORM}" ]; then
320 PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
322 cat > ${PLATFORMTMP} <<PLATFORMHEAD
323 <?xml version='1.0'?>
324 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
325 <platform version="4.1">
326 <zone id="AS0" routing="Full">
330 while [ $i -gt 0 ]; do
331 echo " <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
332 echo " <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
333 echo " <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
338 while [ $i -gt 0 ]; do
340 while [ $j -gt 0 ]; do
341 if [ $i -eq $j ]; then
342 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
344 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
351 cat >> ${PLATFORMTMP} <<PLATFORMFOOT
357 PLATFORMTMP=${PLATFORM}
359 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
360 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
361 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
362 #APPLICATIONTMP="app.xml"
364 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
365 <?xml version='1.0'?>
366 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
367 <platform version="4.1">
370 ##---- cache hostnames of hostfile---------------
371 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
372 hostnames=$(cat ${HOSTFILE} | tr '\n\r' ' ')
375 if [ "${EXTOPT}" = "smpi_replay" ]; then
376 APP_TRACES=$PROC_ARGS
377 if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
378 hosttraces=$(cat ${APP_TRACES} | tr '\n\r' ' ' )
379 NUMTRACES=$(cat ${APP_TRACES} | wc -l)
382 printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
387 ##----------------------------------------------------------
388 ## generate application.xml with hostnames from hostfile:
389 ## the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
390 ## hostfile has less than i lines.
391 ##----------------------------------------------------------
393 HAVE_SEQ="$(which seq 2>/dev/null)"
395 if [ -n "${HAVE_SEQ}" ]; then
396 SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
399 while [ $cnt -lt ${NUMPROCS} ] ; do
407 ##---- generate <actor> tags------------------------------
408 #prepare arguments at once
409 for ARG in $PROC_ARGS; do
410 XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
416 j=$(( i % hostfile_procs + 1 ))
417 host=$(eval "echo \${$j}")
419 ##---- optional display of ranks to actor mapping
420 if [ ${MAPOPT} = 1 ]; then
421 echo "[rank $i] -> $host"
424 echo " <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
425 <argument value=\"smpirun\"/> <!-- instance -->
426 <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
427 if [ ${REPLAY} = 1 ]; then
428 if [ ${NUMTRACES} -gt 1 ]; then
429 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
431 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
434 echo ${XML_ARGS} >> ${APPLICATIONTMP}
436 echo " </actor>" >> ${APPLICATIONTMP}
439 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
442 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
443 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
444 if [ -n "${TRACE_ACTIVE}" ]; then
445 #define trace filename
446 if [ -n "${TRACE_TI_ACTIVE}" ]; then
447 if [ -z "${TRACE_FILENAME}" ]; then
448 TRACE_FILENAME="smpi_simgrid.txt"
450 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
452 if [ -z "${TRACE_FILENAME}" ]; then
453 TRACE_FILENAME="smpi_simgrid.trace"
455 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
458 if [ -n "${TRACE_COMMENT}" ]; then
459 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
462 if [ -n "${TRACE_COMMENT_FILE}" ]; then
463 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
466 if [ -n "${TRACE_GROUPED}" ]; then
467 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
470 if [ -n "${TRACE_RESOURCE}" ]; then
471 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
474 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
476 export SMPI_GLOBAL_SIZE=${NUMPROCS}
477 if [ -n "${KEEP}" ] ; then
478 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
479 if [ ${HOSTFILETMP} = 1 ] ; then
480 echo "Generated hostfile ${HOSTFILE} kept."
482 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
483 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
487 # Execute the process
489 # The shell still need to be alive for the duration in order to do some cleanup after the process.
491 # We are going through great lengths in order to both keep stdin and be able to handle signals:
493 # * The job is launched in the background in order to be able to handle signals.
495 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
496 # is launched in the background: this can be overriden in bash but not in standard bourne shell.
498 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
503 # With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit
504 # status of 128. Try again.
505 while test $status -eq 128 && kill -0 $pid 2>/dev/null; do
511 # Keep temporary files on failures to help debugging
513 if [ ${status} -ne 0 ] ; then
514 if [ -z "${KEEP}" ]; then
515 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
516 if [ ${HOSTFILETMP} = 1 ] ; then
517 echo "Generated hostfile ${HOSTFILE} kept."
519 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
520 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
523 echo "Execution failed with code ${status}."