3 # Copyright (c) 2007-2019. 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]
34 $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> -replay <tracefile> [program] [program-options]
36 -keep-temps # don't remove the generated files after execution
37 -wrapper <command> # use command to run the program (e.g. "valgrind" or "gdb --args")
38 -gdb # run within GDB (-wrapper "gdb --args" -keep-temps)
39 -lldb # run within LLDB (-wrapper "lldb --" -keep-temps)
40 -map # display the machine on which each process rank is mapped
41 -np <numprocs> # use that amount of processes from the hostfile.
42 # By default, all processes of the hostfile are used.
43 -no-privatize # Disable the globals privatization, that is activated by default
44 -trace-ti # activate time independant tracing (for replay, default in smpi_simgrid.txt)
45 -trace # activate tracing (Paje, default in smpi_simgrid.trace)
46 -trace-comment <comment> # put a comment on the top of the trace file
47 -trace-comment-file <file> # put file contents on the top of the trace file as comment
48 -trace-grouped # group MPI processes by location
49 -trace-resource # trace resource utilization
50 -trace-file <tracefile> # name of the tracefile (simgrid_smpi.trace)
51 -replay <tracefile> # replays a trace instead of actually executing an application
53 -version # Displays the SimGrid version (human readable)
54 -git-version # Displays the git hash of SimGrid
56 or (deprecated usage):
57 $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
62 #check if we have at least one parameter
77 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
80 printf '[%s] ** error: %s. Aborting.\n' "$(basename $0)" "$*" >&2
86 if [ -z "${KEEP}" ] ; then
87 if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
91 if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
95 if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
96 rm -f ${UNROLLEDHOSTFILE}
99 if [ -n ${APPLICATIONTMP} ]; then
100 rm -f ${APPLICATIONTMP}
110 # Cleanup and kill the child process:
112 if ! [ -z "$pid" ]; then
117 # Raise the same signal again (remove the traps first):
118 trap - $trapped_signals
121 # This should never happen:
126 for s in $trapped_signals; do
127 trap "smpirun_trap $s" $s
137 NETWORK_BANDWIDTH="$2"
146 if [ ! -f "${PLATFORM}" ]; then
147 die "the file '${PLATFORM}' does not exist"
151 "-hostfile" | "-machinefile")
153 if [ ! -f "${HOSTFILE}" ]; then
154 die "the file '${HOSTFILE}' does not exist"
163 PRIVATIZE="--cfg=smpi/privatization:no"
176 TRACE_TI_ACTIVE="true"
183 "-trace-comment-file")
184 TRACE_COMMENT_FILE="$2"
196 TRACE_RESOURCE="true"
201 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
211 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
217 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
220 "-help" | "--help" | "-h")
224 "-version" | "--version" | "-v")
225 printf '%b\n' "$SIMGRID_VERSION"
228 "-git-version" | "--git-version")
229 printf '%b\n' "$SIMGRID_GITHASH"
235 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
240 # Nothing to do, compatibility.
249 if [ -n "${APP_TRACES}" ] ; then
250 if [ $# -eq 0 ] ; then
251 EXEC="@SMPIREPLAYMAIN@"
257 # check if we still have at least one parameter beyond options
260 echo "Error: no program to execute!"
269 # steel --cfg and --logs options
270 while [ $# -gt 0 ]; do
275 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
280 PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
286 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
287 echo "No hostfile nor platform specified."
292 if [ -z "${HOSTFILE}" ] ; then
294 HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
295 perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
296 # put all <cluster tag on its own line.
297 cat ${PLATFORM} | tr '\n' ' ' | sed 's/<cluster/\n<cluster/' | \
298 perl -ne 'if (m/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/s) {
299 my ($pre,$rad,$post)=($1,$2,$3);
300 for my $elm (split(",",$rad)) {
301 if ($elm=~/^([^-]*?)-([^-]*)$/) {
302 for (my $i=$1; $i<=$2;$i++) {
303 print "$pre$i$post\n";
306 print "$pre$elm$post\n";
309 } elsif (/<cluster/) {
310 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.");
311 }' ${PLATFORM} >> ${HOSTFILE}
313 UNROLLEDHOSTFILETMP=0
315 #parse if our lines are terminated by :num_process
316 multiple_processes=$(grep -c ":" $HOSTFILE)
317 if [ "${multiple_processes}" -gt 0 ] ; then
318 UNROLLEDHOSTFILETMP=1
319 UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
320 perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/' ${HOSTFILE} > ${UNROLLEDHOSTFILE}
321 if [ ${HOSTFILETMP} = 1 ] ; then
325 HOSTFILE=$UNROLLEDHOSTFILE
328 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
329 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
330 if [ ${hostfile_procs} = 0 ] ; then
331 die "the hostfile '${HOSTFILE}' is empty"
334 if [ -z "${NUMPROCS}" ] ; then
335 # Use the amount of processes in the hostfile as default value for the -np parameter
336 NUMPROCS=$hostfile_procs
339 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
340 echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
343 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
344 if [ -z "${PLATFORM}" ]; then
345 PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
347 cat > ${PLATFORMTMP} <<PLATFORMHEAD
348 <?xml version='1.0'?>
349 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
350 <platform version="4.1">
351 <zone id="AS0" routing="Full">
355 while [ $i -gt 0 ]; do
356 echo " <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
357 echo " <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
358 echo " <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
363 while [ $i -gt 0 ]; do
365 while [ $j -gt 0 ]; do
366 if [ $i -eq $j ]; then
367 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
369 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
376 cat >> ${PLATFORMTMP} <<PLATFORMFOOT
382 PLATFORMTMP=${PLATFORM}
384 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
385 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
386 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
387 #APPLICATIONTMP="app.xml"
389 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
390 <?xml version='1.0'?>
391 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
392 <platform version="4.1">
395 ##---- cache hostnames of hostfile---------------
396 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
397 hostnames=$(cat ${HOSTFILE} | tr '\n\r' ' ')
400 if [ -n "${APP_TRACES}" ]; then
401 if [ -f "${APP_TRACES}" ]; then
402 hosttraces=$(cat ${APP_TRACES} | tr '\n\r' ' ' )
403 NUMTRACES=$(cat ${APP_TRACES} | wc -l)
406 printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
411 ##----------------------------------------------------------
412 ## generate application.xml with hostnames from hostfile:
413 ## the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
414 ## hostfile has less than i lines.
415 ##----------------------------------------------------------
417 HAVE_SEQ="$(which seq 2>/dev/null)"
419 if [ -n "${HAVE_SEQ}" ]; then
420 SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
423 while [ $cnt -lt ${NUMPROCS} ] ; do
431 ##---- generate <actor> tags------------------------------
432 #prepare arguments at once
433 for ARG in $PROC_ARGS; do
434 XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
440 j=$(( i % hostfile_procs + 1 ))
441 host=$(eval "echo \${$j}")
443 ##---- optional display of ranks to actor mapping
444 if [ ${MAPOPT} = 1 ]; then
445 echo "[rank $i] -> $host"
448 echo " <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
449 <prop id=\"instance_id\" value=\"smpirun\"/>
450 <prop id=\"rank\" value=\"$i\"/>" >> ${APPLICATIONTMP}
451 if [ ${REPLAY} = 1 ]; then
452 echo " <prop id=\"smpi_replay\" value=\"true\"/>" >> ${APPLICATIONTMP}
453 if [ ${NUMTRACES} -gt 1 ]; then
454 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
456 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
459 echo ${XML_ARGS} >> ${APPLICATIONTMP}
461 echo " </actor>" >> ${APPLICATIONTMP}
464 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
467 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
468 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
469 if [ -n "${TRACE_ACTIVE}" ]; then
470 #define trace filename
471 if [ -n "${TRACE_TI_ACTIVE}" ]; then
472 if [ -z "${TRACE_FILENAME}" ]; then
473 TRACE_FILENAME="smpi_simgrid.txt"
475 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
477 if [ -z "${TRACE_FILENAME}" ]; then
478 TRACE_FILENAME="smpi_simgrid.trace"
480 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
483 if [ -n "${TRACE_COMMENT}" ]; then
484 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
487 if [ -n "${TRACE_COMMENT_FILE}" ]; then
488 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
491 if [ -n "${TRACE_GROUPED}" ]; then
492 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
495 if [ -n "${TRACE_RESOURCE}" ]; then
496 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
499 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
501 export SMPI_GLOBAL_SIZE=${NUMPROCS}
502 if [ -n "${KEEP}" ] ; then
503 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
504 if [ ${HOSTFILETMP} = 1 ] ; then
505 echo "Generated hostfile ${HOSTFILE} kept."
507 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
508 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
512 # Execute the process
514 # The shell still need to be alive for the duration in order to do some cleanup after the process.
516 # We are going through great lengths in order to both keep stdin and be able to handle signals:
518 # * The job is launched in the background in order to be able to handle signals.
520 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
521 # is launched in the background: this can be overriden in bash but not in standard bourne shell.
523 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
528 # With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit
529 # status of 128. Try again.
530 while test $status -eq 128 && kill -0 $pid 2>/dev/null; do
536 # Keep temporary files on failures to help debugging
538 if [ ${status} -ne 0 ] ; then
539 if [ -z "${KEEP}" ]; then
540 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
541 if [ ${HOSTFILETMP} = 1 ] ; then
542 echo "Generated hostfile ${HOSTFILE} kept."
544 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
545 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
548 echo "Execution failed with code ${status}."