src/smpi/smpirun.in

   1 #! /bin/sh
   2
   3 # Copyright (c) 2007-2015. The SimGrid Team.
   4 # All rights reserved.
   5
   6 # This program is free software; you can redistribute it and/or modify it
   7 # under the terms of the license (GNU LGPL) which comes with this package.
   8
   9 @CMAKE_SMPI_COMMAND@
  10
  11 SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
  12 SIMGRID_GITHASH="@SIMGRID_GITHASH@"
  13
  14 DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps"
  15 DEFAULT_LOOPBACK_LATENCY="0.000004s"
  16 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))Bps"
  17 DEFAULT_NETWORK_LATENCY="0.000005s"
  18 DEFAULT_NUMPROCS="4"
  19 DEFAULT_SPEED="100flops"
  20
  21 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
  22 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
  23 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
  24 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
  25 SPEED="${DEFAULT_SPEED}"
  26
  27 SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP-gamma:4194304"
  28
  29 #usage to print the way this script should be called
  30 usage () {
  31     cat <<EOF
  32 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
  33 Options:
  34   -keep-temps                # don't remove the generated files after execution
  35   -wrapper <command>         # use command to run the program (e.g. "valgrind")
  36   -map                       # display the machine on which each process rank is mapped
  37   -np <numprocs>             # use that amount of processes from the hostfile.
  38                              # By default, all processes of the hostfile are used.
  39   -trace-ti                  # activate time independant tracing (for replay, default in smpi_simgrid.txt)
  40   -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
  41   -trace-comment <comment>   # put a comment on the top of the trace file
  42   -trace-comment-file <file> # put file contents on the top of the trace file as comment
  43   -trace-grouped             # group MPI processes by location
  44   -trace-resource            # trace resource utilization
  45   -trace-viva                # generate configuration for Viva's GraphView
  46   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  47   -ext <value>               # additional parameter (reserved)
  48
  49   -version                   # Displays the SimGrid version (human readable)
  50   -git-version               # Displays the git hash of SimGrid
  51
  52 or (deprecated usage):
  53   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  54
  55 EOF
  56 }
  57
  58 #check if we have at least one parameter
  59 if [ $# -eq 0 ]
  60 then
  61     usage
  62     exit
  63 fi
  64
  65 EXTOPT=""
  66 WRAPPER=""
  67 HOSTFILE=""
  68 HOSTFILETMP=0
  69
  70 unset pid
  71
  72 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
  73
  74 smpirun_cleanup()
  75 {
  76   if [ -z "${KEEP}" ] ; then
  77       if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
  78         rm -f ${PLATFORMTMP}
  79         PLATFORMTMP=""
  80       fi
  81       if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
  82           rm -f ${HOSTFILE}
  83           HOSTFILE=""
  84       fi
  85       if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
  86           rm -f ${UNROLLEDHOSTFILE}
  87           UNROLLEDHOSTFILE=""
  88       fi
  89       if [ -n ${APPLICATIONTMP} ]; then
  90         rm -f ${APPLICATIONTMP}
  91         APPLICATIONTMP=""
  92       fi
  93   fi
  94 }
  95
  96 smpirun_trap() {
  97   local sig
  98   sig="$1"
  99
 100   # Cleanup and kill the child process:
 101   smpirun_cleanup
 102   if ! [ -z "$pid" ]; then
 103     kill -TERM $pid
 104   fi
 105   unset pid
 106
 107   # Raise the same signal again (remove the traps first):
 108   trap - $trapped_signals
 109   kill -$sig $$
 110
 111   # This should never happen:
 112   kill -ABRT $$
 113   kill -TERM $$
 114 }
 115
 116 for s in $trapped_signals; do
 117   trap "smpirun_trap $s" $s
 118 done
 119
 120 while true; do
 121     case "$1" in
 122         "-np" | "-n")
 123             NUMPROCS="$2"
 124             shift 2
 125             ;;
 126         "-bandwidth")
 127             NETWORK_BANDWIDTH="$2"
 128             shift 2
 129             ;;
 130         "-latency")
 131             NETWORK_LATENCY="$2"
 132             shift 2
 133             ;;
 134         "-platform")
 135             PLATFORM="$2"
 136             if [ ! -f "${PLATFORM}" ]; then
 137                 echo "[`basename $0`] ** error: the file '${PLATFORM}' does not exist. Aborting."
 138                 exit 1
 139             fi
 140             shift 2
 141             ;;
 142         "-hostfile")
 143             HOSTFILE="$2"
 144             if [ ! -f "${HOSTFILE}" ]; then
 145                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 146                 exit 1
 147             fi
 148             shift 2
 149             ;;
 150         "-machinefile")
 151             HOSTFILE="$2"
 152             if [ ! -f "${HOSTFILE}" ]; then
 153                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 154                 exit 1
 155             fi
 156             shift 2
 157             ;;
 158         "-ext")
 159             EXTOPT="$2"
 160             shift 2
 161             ;;
 162         "-map")
 163             MAPOPT="true"
 164             shift 1
 165             ;;
 166         "-trace")
 167             TRACE_ACTIVE="true"
 168             shift 1
 169             ;;
 170         "-trace-ti")
 171             TRACE_ACTIVE="true"
 172             TRACE_TI_ACTIVE="true"
 173             shift 1
 174             ;;
 175         "-trace-comment")
 176             TRACE_COMMENT="$2"
 177             shift 2
 178             ;;
 179         "-trace-comment-file")
 180             TRACE_COMMENT_FILE="$2"
 181             shift 2
 182             ;;
 183         "-trace-file")
 184             TRACE_FILENAME="$2"
 185             shift 2
 186             ;;
 187         "-trace-grouped")
 188             TRACE_GROUPED="true"
 189             shift 1
 190             ;;
 191         "-trace-resource")
 192             TRACE_RESOURCE="true"
 193             shift 1
 194             ;;
 195         "-trace-viva")
 196             TRACE_VIVA="true"
 197             shift 1
 198             ;;
 199         "-keep-temps")
 200             KEEP="true"
 201             shift 1
 202             ;;
 203         "-wrapper")
 204             WRAPPER="$2"
 205             shift 2
 206             ;;
 207         "-help" | "--help" | "-h")
 208             usage
 209             exit 0
 210             ;;
 211         "-version" | "--version" | "-v")
 212             printf '%b\n' "$SIMGRID_VERSION"
 213             exit 0
 214             ;;
 215         "-git-version" | "--git-version")
 216             printf '%b\n' "$SIMGRID_GITHASH"
 217             exit 0
 218             ;;
 219         "--cfg="*|"--log="*)
 220             for OPT in ${1#*=}
 221             do
 222                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 223             done
 224             shift 1
 225             ;;
 226         "-foreground")
 227             # Nothing to do, compatibility.
 228             shift 1
 229             ;;
 230         *)
 231             break
 232             ;;
 233     esac
 234 done
 235
 236 if [ -n "$WRAPPER" ]; then
 237     EXEC="$WRAPPER $1"
 238 else
 239     EXEC="$1"
 240 fi
 241 shift
 242
 243 # steel --cfg and --logs options
 244 while [ $# -gt 0 ]; do
 245     case "$1" in
 246         "--cfg="*|"--log="*)
 247             for OPT in ${1#*=}
 248             do
 249                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 250             done
 251             shift 1
 252             ;;
 253         *)
 254             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 255             shift
 256             ;;
 257     esac
 258 done
 259
 260 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 261     echo "No hostfile nor platform specified."
 262     usage
 263     exit 1
 264 fi
 265
 266 if [ -z "${HOSTFILE}" ] ; then
 267     HOSTFILETMP=1
 268     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 269     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 270     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 271                 my ($pre,$rad,$post)=($1,$2,$3);
 272                 for my $elm (split(",",$rad)) {
 273                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 274                      for (my $i=$1; $i<=$2;$i++) {
 275                         print "$pre$i$post\n";
 276                      }
 277                   } else {
 278                      print "$pre$elm$post\n";
 279                   }
 280                 }
 281               } elsif (/<cluster/) {
 282                 die ("Unparsable cluster tag. Either provide an hostfile yourself or give the attributes prefix, radical and suffix in that order on the <cluster line");
 283               }' ${PLATFORM} >> ${HOSTFILE}
 284 fi
 285 UNROLLEDHOSTFILETMP=0
 286
 287 #parse if our lines are terminated by :num_process
 288 multiple_processes=`grep -c ":" $HOSTFILE`
 289 if [ "${multiple_processes}" -gt 0 ] ; then
 290     UNROLLEDHOSTFILETMP=1
 291     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 292     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 293     if [ ${HOSTFILETMP} = 1 ] ; then
 294         rm ${HOSTFILE}
 295         HOSTFILETMP=0
 296     fi
 297     HOSTFILE=$UNROLLEDHOSTFILE
 298 fi
 299
 300 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 301 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 302 if [ ${hostfile_procs} = 0 ] ; then
 303    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
 304    exit 1
 305 fi
 306
 307 if [ -z "${NUMPROCS}" ] ; then
 308     # Use the amount of processes in the hostfile as default value for the -np parameter
 309     NUMPROCS=$hostfile_procs
 310 fi
 311
 312 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 313     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 314 fi
 315
 316 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 317 if [ -z "${PLATFORM}" ]; then
 318     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 319
 320     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 321 <?xml version='1.0'?>
 322 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 323 <platform version="4">
 324 <AS id="AS0" routing="Full">
 325 PLATFORMHEAD
 326
 327     i=${NUMPROCS}
 328     while [ $i -gt 0 ]; do
 329         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 330         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 331         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 332         i=$((i - 1))
 333     done
 334
 335     i=${NUMPROCS}
 336     while [ $i -gt 0 ]; do
 337         j=${NUMPROCS}
 338         while [ $j -gt 0 ]; do
 339             if [ $i -eq $j ]; then
 340                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 341             else
 342                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 343             fi
 344             j=$((j - 1))
 345         done
 346         i=$((i - 1))
 347     done
 348
 349     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 350 </AS>
 351 </platform>
 352 PLATFORMFOOT
 353
 354 else
 355     PLATFORMTMP=${PLATFORM}
 356 fi
 357 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 358 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 359 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 360 #APPLICATIONTMP="app.xml"
 361
 362 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 363 <?xml version='1.0'?>
 364 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 365 <platform version="4">
 366 APPLICATIONHEAD
 367
 368 ##---- cache hostnames of hostfile---------------
 369 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 370     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 371     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
 372 fi
 373
 374 if [ "${EXTOPT}" = "smpi_replay" ]; then
 375     APP_TRACES=$PROC_ARGS
 376     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 377         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 378         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 379     else
 380         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 381         exit 1
 382     fi
 383 fi
 384
 385 ##----------------------------------------------------------
 386 ##  generate application.xml with hostnames from hostfile:
 387 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 388 ##  hostfile has less than i lines.
 389 ##----------------------------------------------------------
 390
 391 HAVE_SEQ="`which seq 2>/dev/null`"
 392
 393 if [ -n "${HAVE_SEQ}" ]; then
 394     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 395 else
 396     cnt=0
 397     while [ $cnt -lt ${NUMPROCS} ] ; do
 398         SEQ="$SEQ $cnt"
 399         cnt=$((cnt + 1));
 400     done
 401 fi
 402
 403 ##---- generate <process> tags------------------------------
 404 for i in ${SEQ}
 405 do
 406     if [ -n "${HOSTFILE}" ]; then
 407         j=$(( $i % ${NUMHOSTS} + 1 ))
 408     fi
 409     ##---- optional display of ranks to process mapping
 410     if [ -n "${MAPOPT}" ]; then
 411         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
 412     fi
 413
 414     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
 415         host="host"$($j)
 416     else
 417         host="$(echo $hostnames|cut -d' ' -f$j)"
 418     fi
 419     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
 420     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
 421     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 422     if [ "${EXTOPT}" = "smpi_replay" ]; then
 423         if  [ ${NUMTRACES} -gt 1 ]; then
 424             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 425         else
 426             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 427         fi
 428     else
 429         for ARG in $PROC_ARGS; do
 430             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
 431         done
 432     fi
 433     echo "  </process>" >> ${APPLICATIONTMP}
 434 done
 435
 436 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 437 </platform>
 438 APPLICATIONFOOT
 439 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 440 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 441 if [ -n "${TRACE_ACTIVE}" ]; then
 442     #define trace filename
 443     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 444         if [ -z "${TRACE_FILENAME}" ]; then
 445             TRACE_FILENAME="smpi_simgrid.txt"
 446         fi
 447         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 448     else
 449         if [ -z "${TRACE_FILENAME}" ]; then
 450             TRACE_FILENAME="smpi_simgrid.trace"
 451         fi
 452         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 453     fi
 454
 455     if [ -n "${TRACE_COMMENT}" ]; then
 456         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 457     fi
 458
 459     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 460         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 461     fi
 462
 463     if [ -n "${TRACE_GROUPED}" ]; then
 464         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 465     fi
 466
 467     if [ -n "${TRACE_RESOURCE}" ]; then
 468         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 469     fi
 470
 471     if [ -n "${TRACE_VIVA}" ]; then
 472         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 473     fi
 474 fi
 475 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 476
 477 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 478 if [ -n "${KEEP}" ] ; then
 479     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 480     if [ ${HOSTFILETMP} = 1 ] ; then
 481         echo "Generated hostfile ${HOSTFILE} kept."
 482     fi
 483     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 484         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 485     fi
 486 fi
 487
 488 # Execute the process
 489 #
 490 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 491 #
 492 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 493 #
 494 # * The job is launched in the background in order to be able to handle signals.
 495 #
 496 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 497 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 498 exec 3<&0
 499 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 500 pid=$!
 501 exec 3>&-
 502 wait $pid
 503 status=$?
 504 pid=""
 505
 506 smpirun_cleanup
 507
 508 exit $status