3 # Copyright (c) 2007-2020. 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 -vgdb # run within Valgrind+GDB (-wrapper "valgrind --vgdb=yes --vgdb-error=0" -keep-temps)
41 -map # display the machine on which each process rank is mapped
42 -np <numprocs> # use that amount of processes from the hostfile.
43 # By default, all processes of the hostfile are used.
44 -no-privatize # Disable the globals privatization, that is activated by default
45 -trace-ti # activate time independent tracing (for replay, default in smpi_simgrid.txt)
46 -trace # activate tracing (Paje, default in smpi_simgrid.trace)
47 -trace-comment <comment> # put a comment on the top of the trace file
48 -trace-comment-file <file> # put file contents on the top of the trace file as comment
49 -trace-grouped # group MPI processes by location
50 -trace-resource # trace resource utilization
51 -trace-file <tracefile> # name of the tracefile (simgrid_smpi.trace)
52 -replay <tracefile> # replays a trace instead of actually executing an application
54 -version # Displays the SimGrid version (human readable)
55 -git-version # Displays the git hash of SimGrid
57 or (deprecated usage):
58 $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
63 #check if we have at least one parameter
79 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
82 printf '[%s] ** error: %s. Aborting.\n' "$(basename $0)" "$*" >&2
88 if [ -z "${KEEP}" ] ; then
89 if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
93 if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
97 if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
98 rm -f ${UNROLLEDHOSTFILE}
101 if [ -n ${APPLICATIONTMP} ]; then
102 rm -f ${APPLICATIONTMP}
112 # Cleanup and kill the child process:
114 if ! [ -z "$pid" ]; then
119 # Raise the same signal again (remove the traps first):
120 trap - $trapped_signals
123 # This should never happen:
128 for s in $trapped_signals; do
129 trap "smpirun_trap $s" $s
139 NETWORK_BANDWIDTH="$2"
148 if [ ! -f "${PLATFORM}" ]; then
149 die "the file '${PLATFORM}' does not exist"
153 "-hostfile" | "-machinefile")
155 if [ ! -f "${HOSTFILE}" ]; then
156 die "the file '${HOSTFILE}' does not exist"
165 PRIVATIZE="--cfg=smpi/privatization:no"
178 TRACE_TI_ACTIVE="true"
185 "-trace-comment-file")
186 TRACE_COMMENT_FILE="$2"
198 TRACE_RESOURCE="true"
203 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
217 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
221 WRAPPER="valgrind --vgdb=yes --vgdb-error=0"
223 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
229 SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
232 "-help" | "--help" | "-h")
236 "-version" | "--version" | "-v")
237 printf '%b\n' "$SIMGRID_VERSION"
240 "-git-version" | "--git-version")
241 printf '%b\n' "$SIMGRID_GITHASH"
247 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
252 # Nothing to do, compatibility.
261 if [ -n "${APP_TRACES}" ] ; then
262 if [ $# -eq 0 ] ; then
263 EXEC="@SMPIREPLAYMAIN@"
269 # check if we still have at least one parameter beyond options
272 echo "Error: no program to execute!"
281 # steal --cfg and --logs options
282 while [ $# -gt 0 ]; do
287 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
292 PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
298 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
299 echo "No hostfile nor platform specified."
304 if [ -z "${HOSTFILE}" ] ; then
306 HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
307 @PYTHON_EXECUTABLE@ -c '
308 import xml.etree.ElementTree as ET
312 tree = ET.parse(sys.stdin)
314 for elem in tree.findall(".//host"):
315 print(elem.attrib["id"])
317 for elem in tree.findall(".//cluster"):
318 prefix = elem.attrib["prefix"]
319 radical = elem.attrib["radical"]
320 suffix = elem.attrib["suffix"]
321 for r in radical.split(","):
322 m = re.match("^([^-]*?)-([^-]*)$", r)
324 for i in range(int(m.group(1)), int(m.group(2))):
325 print(prefix + str(i) + suffix)
327 print(prefix + r + suffix)
328 ' < ${PLATFORM} > ${HOSTFILE}
330 UNROLLEDHOSTFILETMP=0
332 # parse if our lines are terminated by :num_process
333 if grep -q ':' $HOSTFILE ; then
334 UNROLLEDHOSTFILETMP=1
335 UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
336 @PYTHON_EXECUTABLE@ -c '
340 for line in sys.stdin:
341 m = re.match("(.*):(.*)", line)
343 for i in range(0, int(m.group(2))):
347 ' < ${HOSTFILE} > ${UNROLLEDHOSTFILE}
348 if [ ${HOSTFILETMP} = 1 ] ; then
352 HOSTFILE=$UNROLLEDHOSTFILE
355 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
356 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
357 if [ ${hostfile_procs} = 0 ] ; then
358 die "the hostfile '${HOSTFILE}' is empty"
361 if [ -z "${NUMPROCS}" ] ; then
362 # Use the amount of processes in the hostfile as default value for the -np parameter
363 NUMPROCS=$hostfile_procs
366 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
367 echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
370 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
371 if [ -z "${PLATFORM}" ]; then
372 PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
374 cat > ${PLATFORMTMP} <<PLATFORMHEAD
375 <?xml version='1.0'?>
376 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
377 <platform version="4.1">
378 <zone id="AS0" routing="Full">
382 while [ $i -gt 0 ]; do
383 echo " <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
384 echo " <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
385 echo " <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
390 while [ $i -gt 0 ]; do
392 while [ $j -gt 0 ]; do
393 if [ $i -eq $j ]; then
394 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
396 echo " <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
403 cat >> ${PLATFORMTMP} <<PLATFORMFOOT
409 PLATFORMTMP=${PLATFORM}
411 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
412 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
413 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
414 #APPLICATIONTMP="app.xml"
416 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
417 <?xml version='1.0'?>
418 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
419 <platform version="4.1">
422 ##---- cache hostnames of hostfile---------------
423 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
424 hostnames=$(cat ${HOSTFILE} | tr '\n\r' ' ')
427 if [ -n "${APP_TRACES}" ]; then
428 if [ -f "${APP_TRACES}" ]; then
429 hosttraces=$(cat ${APP_TRACES} | tr '\n\r' ' ' )
430 NUMTRACES=$(cat ${APP_TRACES} | wc -l)
433 printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
438 ##----------------------------------------------------------
439 ## generate application.xml with hostnames from hostfile:
440 ## the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
441 ## hostfile has less than i lines.
442 ##----------------------------------------------------------
444 HAVE_SEQ="$(which seq 2>/dev/null)"
446 if [ -n "${HAVE_SEQ}" ]; then
447 SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
450 while [ $cnt -lt ${NUMPROCS} ] ; do
458 ##---- generate <actor> tags------------------------------
459 #prepare arguments at once
460 for ARG in $PROC_ARGS; do
461 XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
467 j=$(( i % hostfile_procs + 1 ))
468 host=$(eval "echo \${$j}")
470 ##---- optional display of ranks to actor mapping
471 if [ ${MAPOPT} = 1 ]; then
472 echo "[rank $i] -> $host"
475 echo " <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
476 <prop id=\"instance_id\" value=\"smpirun\"/>
477 <prop id=\"rank\" value=\"$i\"/>" >> ${APPLICATIONTMP}
478 if [ ${REPLAY} = 1 ]; then
479 echo " <prop id=\"smpi_replay\" value=\"true\"/>" >> ${APPLICATIONTMP}
480 if [ ${NUMTRACES} -gt 1 ]; then
481 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
483 echo " <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
486 echo ${XML_ARGS} >> ${APPLICATIONTMP}
488 echo " </actor>" >> ${APPLICATIONTMP}
491 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
494 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
495 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
496 if [ -n "${TRACE_ACTIVE}" ]; then
497 #define trace filename
498 if [ -n "${TRACE_TI_ACTIVE}" ]; then
499 if [ -z "${TRACE_FILENAME}" ]; then
500 TRACE_FILENAME="smpi_simgrid.txt"
502 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
504 if [ -z "${TRACE_FILENAME}" ]; then
505 TRACE_FILENAME="smpi_simgrid.trace"
507 TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
510 if [ -n "${TRACE_COMMENT}" ]; then
511 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
514 if [ -n "${TRACE_COMMENT_FILE}" ]; then
515 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
518 if [ -n "${TRACE_GROUPED}" ]; then
519 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
522 if [ -n "${TRACE_RESOURCE}" ]; then
523 TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
526 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
528 # Do not remove, this variable may be used by user code (e.g. StarPU)
529 export SMPI_GLOBAL_SIZE=${NUMPROCS}
530 if [ -n "${KEEP}" -a -z "${QUIET}" ] ; then
531 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
532 if [ ${HOSTFILETMP} = 1 ] ; then
533 echo "Generated hostfile ${HOSTFILE} kept."
535 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
536 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
540 # Execute the process
542 # The shell still need to be alive for the duration in order to do some cleanup after the process.
544 # We are going through great lengths in order to both keep stdin and be able to handle signals:
546 # * The job is launched in the background in order to be able to handle signals.
548 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
549 # is launched in the background: this can be overridden in bash but not in standard bourne shell.
551 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
556 # With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit
557 # status of 128. Try again.
558 while test $status -eq 128 && kill -0 $pid 2>/dev/null; do
564 # Keep temporary files on failures to help debugging
566 if [ ${status} -ne 0 ] ; then
567 if [ -z "${KEEP}" -a -z "${QUIET}" ]; then
568 echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
569 if [ ${HOSTFILETMP} = 1 ] ; then
570 echo "Generated hostfile ${HOSTFILE} kept."
572 if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
573 echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
577 echo "Execution failed with code ${status}."