src/smpi/smpirun.in

   1 #! /bin/sh
   2
   3 # Copyright (c) 2007-2016, The SimGrid Team. All rights reserved.
   4
   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.
   7
   8 @CMAKE_SMPI_COMMAND@
   9
  10 SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
  11 SIMGRID_GITHASH="@SIMGRID_GITHASH@"
  12
  13 DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps"
  14 DEFAULT_LOOPBACK_LATENCY="0.000004s"
  15 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))Bps"
  16 DEFAULT_NETWORK_LATENCY="0.000005s"
  17 DEFAULT_NUMPROCS="4"
  18 DEFAULT_SPEED="100flops"
  19
  20 LOOPBACK_BANDWIDTH="${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}"
  25
  26 PRIVATIZE="--cfg=smpi/privatization:@HAVE_PRIVATIZATION@"
  27
  28 SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP-gamma:4194304"
  29
  30 #usage to print the way this script should be called
  31 usage () {
  32     cat <<EOF
  33 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
  34 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-viva                # generate configuration for Viva's GraphView
  48   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  49   -ext <value>               # additional parameter (reserved)
  50
  51   -version                   # Displays the SimGrid version (human readable)
  52   -git-version               # Displays the git hash of SimGrid
  53
  54 or (deprecated usage):
  55   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  56
  57 EOF
  58 }
  59
  60 #check if we have at least one parameter
  61 if [ $# -eq 0 ]
  62 then
  63     usage
  64     exit
  65 fi
  66
  67 EXTOPT=""
  68 WRAPPER=""
  69 HOSTFILE=""
  70 HOSTFILETMP=0
  71 MAPOPT=0
  72 REPLAY=0
  73
  74 unset pid
  75
  76 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
  77
  78 smpirun_cleanup()
  79 {
  80   if [ -z "${KEEP}" ] ; then
  81       if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
  82         rm -f ${PLATFORMTMP}
  83         PLATFORMTMP=""
  84       fi
  85       if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
  86           rm -f ${HOSTFILE}
  87           HOSTFILE=""
  88       fi
  89       if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
  90           rm -f ${UNROLLEDHOSTFILE}
  91           UNROLLEDHOSTFILE=""
  92       fi
  93       if [ -n ${APPLICATIONTMP} ]; then
  94         rm -f ${APPLICATIONTMP}
  95         APPLICATIONTMP=""
  96       fi
  97   fi
  98 }
  99
 100 smpirun_trap() {
 101   local sig
 102   sig="$1"
 103
 104   # Cleanup and kill the child process:
 105   smpirun_cleanup
 106   if ! [ -z "$pid" ]; then
 107     kill -TERM $pid
 108   fi
 109   unset pid
 110
 111   # Raise the same signal again (remove the traps first):
 112   trap - $trapped_signals
 113   kill -$sig $$
 114
 115   # This should never happen:
 116   kill -ABRT $$
 117   kill -TERM $$
 118 }
 119
 120 for s in $trapped_signals; do
 121   trap "smpirun_trap $s" $s
 122 done
 123
 124 while true; do
 125     case "$1" in
 126         "-np" | "-n")
 127             NUMPROCS="$2"
 128             shift 2
 129             ;;
 130         "-bandwidth")
 131             NETWORK_BANDWIDTH="$2"
 132             shift 2
 133             ;;
 134         "-latency")
 135             NETWORK_LATENCY="$2"
 136             shift 2
 137             ;;
 138         "-platform")
 139             PLATFORM="$2"
 140             if [ ! -f "${PLATFORM}" ]; then
 141                 echo "[`basename $0`] ** error: the file '${PLATFORM}' does not exist. Aborting."
 142                 exit 1
 143             fi
 144             shift 2
 145             ;;
 146         "-hostfile")
 147             HOSTFILE="$2"
 148             if [ ! -f "${HOSTFILE}" ]; then
 149                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 150                 exit 1
 151             fi
 152             shift 2
 153             ;;
 154         "-machinefile")
 155             HOSTFILE="$2"
 156             if [ ! -f "${HOSTFILE}" ]; then
 157                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 158                 exit 1
 159             fi
 160             shift 2
 161             ;;
 162         "-ext")
 163             EXTOPT="$2"
 164             shift 2
 165             ;;
 166         "-no-privatize")
 167             PRIVATIZE=""
 168             shift 1
 169             ;;
 170         "-map")
 171             MAPOPT=1
 172             shift 1
 173             ;;
 174         "-trace")
 175             TRACE_ACTIVE="true"
 176             shift 1
 177             ;;
 178         "-trace-ti")
 179             TRACE_ACTIVE="true"
 180             TRACE_TI_ACTIVE="true"
 181             shift 1
 182             ;;
 183         "-trace-comment")
 184             TRACE_COMMENT="$2"
 185             shift 2
 186             ;;
 187         "-trace-comment-file")
 188             TRACE_COMMENT_FILE="$2"
 189             shift 2
 190             ;;
 191         "-trace-file")
 192             TRACE_FILENAME="$2"
 193             shift 2
 194             ;;
 195         "-trace-grouped")
 196             TRACE_GROUPED="true"
 197             shift 1
 198             ;;
 199         "-trace-resource")
 200             TRACE_RESOURCE="true"
 201             shift 1
 202             ;;
 203         "-trace-viva")
 204             TRACE_VIVA="true"
 205             shift 1
 206             ;;
 207         "-keep-temps")
 208             KEEP="true"
 209             SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
 210             shift 1
 211             ;;
 212         "-wrapper")
 213             WRAPPER="$2"
 214             shift 2
 215             ;;
 216         "-help" | "--help" | "-h")
 217             usage
 218             exit 0
 219             ;;
 220         "-version" | "--version" | "-v")
 221             printf '%b\n' "$SIMGRID_VERSION"
 222             exit 0
 223             ;;
 224         "-git-version" | "--git-version")
 225             printf '%b\n' "$SIMGRID_GITHASH"
 226             exit 0
 227             ;;
 228         "--cfg="*|"--log="*)
 229             for OPT in ${1#*=}
 230             do
 231                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 232             done
 233             shift 1
 234             ;;
 235         "-foreground")
 236             # Nothing to do, compatibility.
 237             shift 1
 238             ;;
 239         *)
 240             break
 241             ;;
 242     esac
 243 done
 244
 245 # check if we still have at least one parameter beyond options
 246 if [ $# -eq 0 ]
 247 then
 248     echo "Error: no program to execute!"
 249     usage
 250     exit
 251 fi
 252
 253 EXEC="$1"
 254 shift
 255
 256 # steel --cfg and --logs options
 257 while [ $# -gt 0 ]; do
 258     case "$1" in
 259         "--cfg="*|"--log="*)
 260             for OPT in ${1#*=}
 261             do
 262                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 263             done
 264             shift 1
 265             ;;
 266         *)
 267             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 268             shift
 269             ;;
 270     esac
 271 done
 272
 273 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 274     echo "No hostfile nor platform specified."
 275     usage
 276     exit 1
 277 fi
 278
 279 if [ -z "${HOSTFILE}" ] ; then
 280     HOSTFILETMP=1
 281     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 282     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 283     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 284                 my ($pre,$rad,$post)=($1,$2,$3);
 285                 for my $elm (split(",",$rad)) {
 286                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 287                      for (my $i=$1; $i<=$2;$i++) {
 288                         print "$pre$i$post\n";
 289                      }
 290                   } else {
 291                      print "$pre$elm$post\n";
 292                   }
 293                 }
 294               } elsif (/<cluster/) {
 295             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)");
 296               }' ${PLATFORM} >> ${HOSTFILE}
 297 fi
 298 UNROLLEDHOSTFILETMP=0
 299
 300 #parse if our lines are terminated by :num_process
 301 multiple_processes=$(grep -c ":" $HOSTFILE)
 302 if [ "${multiple_processes}" -gt 0 ] ; then
 303     UNROLLEDHOSTFILETMP=1
 304     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 305     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 306     if [ ${HOSTFILETMP} = 1 ] ; then
 307         rm ${HOSTFILE}
 308         HOSTFILETMP=0
 309     fi
 310     HOSTFILE=$UNROLLEDHOSTFILE
 311 fi
 312
 313 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 314 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
 315 if [ ${hostfile_procs} = 0 ] ; then
 316    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
 317    exit 1
 318 fi
 319
 320 if [ -z "${NUMPROCS}" ] ; then
 321     # Use the amount of processes in the hostfile as default value for the -np parameter
 322     NUMPROCS=$hostfile_procs
 323 fi
 324
 325 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 326     echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 327 fi
 328
 329 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 330 if [ -z "${PLATFORM}" ]; then
 331     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 332
 333     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 334 <?xml version='1.0'?>
 335 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 336 <platform version="4.1">
 337 <zone id="AS0" routing="Full">
 338 PLATFORMHEAD
 339
 340     i=${NUMPROCS}
 341     while [ $i -gt 0 ]; do
 342         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 343         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 344         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 345         i=$((i - 1))
 346     done
 347
 348     i=${NUMPROCS}
 349     while [ $i -gt 0 ]; do
 350         j=${NUMPROCS}
 351         while [ $j -gt 0 ]; do
 352             if [ $i -eq $j ]; then
 353                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 354             else
 355                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 356             fi
 357             j=$((j - 1))
 358         done
 359         i=$((i - 1))
 360     done
 361
 362     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 363 </zone>
 364 </platform>
 365 PLATFORMFOOT
 366
 367 else
 368     PLATFORMTMP=${PLATFORM}
 369 fi
 370 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 371 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 372 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 373 #APPLICATIONTMP="app.xml"
 374
 375 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 376 <?xml version='1.0'?>
 377 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 378 <platform version="4.1">
 379 APPLICATIONHEAD
 380
 381 ##---- cache hostnames of hostfile---------------
 382 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 383     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 384 fi
 385
 386 if [ "${EXTOPT}" = "smpi_replay" ]; then
 387     APP_TRACES=$PROC_ARGS
 388     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 389         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 390         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 391         REPLAY=1
 392     else
 393         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 394         exit 1
 395     fi
 396 fi
 397
 398 ##----------------------------------------------------------
 399 ##  generate application.xml with hostnames from hostfile:
 400 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 401 ##  hostfile has less than i lines.
 402 ##----------------------------------------------------------
 403
 404 HAVE_SEQ="$(which seq 2>/dev/null)"
 405
 406 if [ -n "${HAVE_SEQ}" ]; then
 407     SEQ=$(${HAVE_SEQ} 0 $((${NUMPROCS}-1)))
 408 else
 409     cnt=0
 410     while [ $cnt -lt ${NUMPROCS} ] ; do
 411         SEQ="$SEQ $cnt"
 412         cnt=$((cnt + 1));
 413     done
 414 fi
 415
 416 set -- $hostnames
 417
 418 ##---- generate <actor> tags------------------------------
 419 #prepare arguments at once
 420 for ARG in $PROC_ARGS; do
 421   XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
 422 "
 423 done
 424
 425 for i in ${SEQ}
 426 do
 427     j=$(( $i % ${hostfile_procs} + 1 ))
 428     host=$(eval "echo \${$j}")
 429
 430     ##---- optional display of ranks to actor mapping
 431     if [ ${MAPOPT} = 1 ]; then
 432       echo "[rank $i] -> $host"
 433     fi
 434
 435     echo "  <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
 436     <argument value=\"1\"/> <!-- instance -->
 437     <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 438     if [ ${REPLAY} = 1 ]; then
 439         if  [ ${NUMTRACES} -gt 1 ]; then
 440             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 441         else
 442             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 443         fi
 444     else
 445     echo ${XML_ARGS} >> ${APPLICATIONTMP}
 446     fi
 447     echo "  </actor>" >> ${APPLICATIONTMP}
 448 done
 449
 450 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 451 </platform>
 452 APPLICATIONFOOT
 453 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 454 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 455 if [ -n "${TRACE_ACTIVE}" ]; then
 456     #define trace filename
 457     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 458         if [ -z "${TRACE_FILENAME}" ]; then
 459             TRACE_FILENAME="smpi_simgrid.txt"
 460         fi
 461         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 462     else
 463         if [ -z "${TRACE_FILENAME}" ]; then
 464             TRACE_FILENAME="smpi_simgrid.trace"
 465         fi
 466         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 467     fi
 468
 469     if [ -n "${TRACE_COMMENT}" ]; then
 470         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 471     fi
 472
 473     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 474         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 475     fi
 476
 477     if [ -n "${TRACE_GROUPED}" ]; then
 478         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 479     fi
 480
 481     if [ -n "${TRACE_RESOURCE}" ]; then
 482         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 483     fi
 484
 485     if [ -n "${TRACE_VIVA}" ]; then
 486         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 487     fi
 488 fi
 489 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 490
 491 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 492 if [ -n "${KEEP}" ] ; then
 493     echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 494     if [ ${HOSTFILETMP} = 1 ] ; then
 495         echo "Generated hostfile ${HOSTFILE} kept."
 496     fi
 497     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 498         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 499     fi
 500 fi
 501
 502 # Execute the process
 503 #
 504 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 505 #
 506 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 507 #
 508 # * The job is launched in the background in order to be able to handle signals.
 509 #
 510 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 511 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 512 exec 3<&0
 513 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 514 pid=$!
 515 exec 3>&-
 516 wait $pid
 517 status=$?
 518 pid=""
 519
 520 # Keep temporary files on failures to help debugging
 521 #
 522 if [ ${status} -ne 0 ] ; then
 523     if [ -z ${KEEP} ]; then
 524         echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 525         if [ ${HOSTFILETMP} = 1 ] ; then
 526             echo "Generated hostfile ${HOSTFILE} kept."
 527         fi
 528         if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 529             echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 530         fi
 531     fi
 532     echo "Execution failed with code ${status}."
 533     KEEP=true
 534 fi
 535
 536 smpirun_cleanup
 537
 538 exit $status