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/privatize-global-variables:yes"
  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")
  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             shift 1
 210             ;;
 211         "-wrapper")
 212             WRAPPER="$2"
 213             shift 2
 214             ;;
 215         "-help" | "--help" | "-h")
 216             usage
 217             exit 0
 218             ;;
 219         "-version" | "--version" | "-v")
 220             printf '%b\n' "$SIMGRID_VERSION"
 221             exit 0
 222             ;;
 223         "-git-version" | "--git-version")
 224             printf '%b\n' "$SIMGRID_GITHASH"
 225             exit 0
 226             ;;
 227         "--cfg="*|"--log="*)
 228             for OPT in ${1#*=}
 229             do
 230                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 231             done
 232             shift 1
 233             ;;
 234         "-foreground")
 235             # Nothing to do, compatibility.
 236             shift 1
 237             ;;
 238         *)
 239             break
 240             ;;
 241     esac
 242 done
 243
 244 # check if we still have at least one parameter beyond options
 245 if [ $# -eq 0 ]
 246 then
 247     echo "Error: no program to execute!"
 248     usage
 249     exit
 250 fi
 251
 252 if [ -n "$WRAPPER" ]; then
 253     EXEC="$WRAPPER $1"
 254 else
 255     EXEC="$1"
 256 fi
 257 shift
 258
 259 # steel --cfg and --logs options
 260 while [ $# -gt 0 ]; do
 261     case "$1" in
 262         "--cfg="*|"--log="*)
 263             for OPT in ${1#*=}
 264             do
 265                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 266             done
 267             shift 1
 268             ;;
 269         *)
 270             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 271             shift
 272             ;;
 273     esac
 274 done
 275
 276 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 277     echo "No hostfile nor platform specified."
 278     usage
 279     exit 1
 280 fi
 281
 282 if [ -z "${HOSTFILE}" ] ; then
 283     HOSTFILETMP=1
 284     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 285     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 286     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 287                 my ($pre,$rad,$post)=($1,$2,$3);
 288                 for my $elm (split(",",$rad)) {
 289                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 290                      for (my $i=$1; $i<=$2;$i++) {
 291                         print "$pre$i$post\n";
 292                      }
 293                   } else {
 294                      print "$pre$elm$post\n";
 295                   }
 296                 }
 297               } elsif (/<cluster/) {
 298             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)");
 299               }' ${PLATFORM} >> ${HOSTFILE}
 300 fi
 301 UNROLLEDHOSTFILETMP=0
 302
 303 #parse if our lines are terminated by :num_process
 304 multiple_processes=`grep -c ":" $HOSTFILE`
 305 if [ "${multiple_processes}" -gt 0 ] ; then
 306     UNROLLEDHOSTFILETMP=1
 307     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 308     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 309     if [ ${HOSTFILETMP} = 1 ] ; then
 310         rm ${HOSTFILE}
 311         HOSTFILETMP=0
 312     fi
 313     HOSTFILE=$UNROLLEDHOSTFILE
 314 fi
 315
 316 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 317 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 318 if [ ${hostfile_procs} = 0 ] ; then
 319    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
 320    exit 1
 321 fi
 322
 323 if [ -z "${NUMPROCS}" ] ; then
 324     # Use the amount of processes in the hostfile as default value for the -np parameter
 325     NUMPROCS=$hostfile_procs
 326 fi
 327
 328 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 329     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 330 fi
 331
 332 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 333 if [ -z "${PLATFORM}" ]; then
 334     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 335
 336     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 337 <?xml version='1.0'?>
 338 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 339 <platform version="4">
 340 <AS id="AS0" routing="Full">
 341 PLATFORMHEAD
 342
 343     i=${NUMPROCS}
 344     while [ $i -gt 0 ]; do
 345         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 346         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 347         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 348         i=$((i - 1))
 349     done
 350
 351     i=${NUMPROCS}
 352     while [ $i -gt 0 ]; do
 353         j=${NUMPROCS}
 354         while [ $j -gt 0 ]; do
 355             if [ $i -eq $j ]; then
 356                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 357             else
 358                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 359             fi
 360             j=$((j - 1))
 361         done
 362         i=$((i - 1))
 363     done
 364
 365     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 366 </AS>
 367 </platform>
 368 PLATFORMFOOT
 369
 370 else
 371     PLATFORMTMP=${PLATFORM}
 372 fi
 373 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 374 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 375 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 376 #APPLICATIONTMP="app.xml"
 377
 378 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 379 <?xml version='1.0'?>
 380 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 381 <platform version="4">
 382 APPLICATIONHEAD
 383
 384 ##---- cache hostnames of hostfile---------------
 385 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 386     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 387 fi
 388
 389 if [ "${EXTOPT}" = "smpi_replay" ]; then
 390     APP_TRACES=$PROC_ARGS
 391     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 392         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 393         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 394         REPLAY=1
 395     else
 396         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 397         exit 1
 398     fi
 399 fi
 400
 401 ##----------------------------------------------------------
 402 ##  generate application.xml with hostnames from hostfile:
 403 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 404 ##  hostfile has less than i lines.
 405 ##----------------------------------------------------------
 406
 407 HAVE_SEQ="`which seq 2>/dev/null`"
 408
 409 if [ -n "${HAVE_SEQ}" ]; then
 410     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 411 else
 412     cnt=0
 413     while [ $cnt -lt ${NUMPROCS} ] ; do
 414         SEQ="$SEQ $cnt"
 415         cnt=$((cnt + 1));
 416     done
 417 fi
 418
 419 set -- $hostnames
 420
 421 ##---- generate <process> tags------------------------------
 422 #prepare arguments at once
 423 for ARG in $PROC_ARGS; do
 424   XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
 425 "
 426 done
 427
 428 for i in ${SEQ}
 429 do
 430     j=$(( $i % ${hostfile_procs} + 1 ))
 431     host=$(eval "echo \${$j}")
 432
 433     ##---- optional display of ranks to process mapping
 434     if [ ${MAPOPT} = 1 ]; then
 435       echo "[rank $i] -> $host"
 436     fi
 437
 438     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
 439     <argument value=\"1\"/> <!-- instance -->
 440     <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 441     if [ ${REPLAY} = 1 ]; then
 442         if  [ ${NUMTRACES} -gt 1 ]; then
 443             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 444         else
 445             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 446         fi
 447     else
 448     echo ${XML_ARGS} >> ${APPLICATIONTMP}
 449     fi
 450     echo "  </process>" >> ${APPLICATIONTMP}
 451 done
 452
 453 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 454 </platform>
 455 APPLICATIONFOOT
 456 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 457 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 458 if [ -n "${TRACE_ACTIVE}" ]; then
 459     #define trace filename
 460     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 461         if [ -z "${TRACE_FILENAME}" ]; then
 462             TRACE_FILENAME="smpi_simgrid.txt"
 463         fi
 464         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 465     else
 466         if [ -z "${TRACE_FILENAME}" ]; then
 467             TRACE_FILENAME="smpi_simgrid.trace"
 468         fi
 469         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 470     fi
 471
 472     if [ -n "${TRACE_COMMENT}" ]; then
 473         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 474     fi
 475
 476     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 477         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 478     fi
 479
 480     if [ -n "${TRACE_GROUPED}" ]; then
 481         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 482     fi
 483
 484     if [ -n "${TRACE_RESOURCE}" ]; then
 485         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 486     fi
 487
 488     if [ -n "${TRACE_VIVA}" ]; then
 489         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 490     fi
 491 fi
 492 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 493
 494 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 495 if [ -n "${KEEP}" ] ; then
 496     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PRIVATIZE} ${PLATFORMTMP} ${APPLICATIONTMP}
 497     if [ ${HOSTFILETMP} = 1 ] ; then
 498         echo "Generated hostfile ${HOSTFILE} kept."
 499     fi
 500     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 501         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 502     fi
 503 fi
 504
 505 # Execute the process
 506 #
 507 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 508 #
 509 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 510 #
 511 # * The job is launched in the background in order to be able to handle signals.
 512 #
 513 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 514 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 515 exec 3<&0
 516 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 517 pid=$!
 518 exec 3>&-
 519 wait $pid
 520 status=$?
 521 pid=""
 522
 523 # Keep temporary files on failures to help debugging
 524 #
 525 if [ ${status} -ne 0 ] ; then
 526     if [ -z ${KEEP} ]; then
 527         echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PRIVATIZE} ${PLATFORMTMP} ${APPLICATIONTMP}
 528         if [ ${HOSTFILETMP} = 1 ] ; then
 529             echo "Generated hostfile ${HOSTFILE} kept."
 530         fi
 531         if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 532             echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 533         fi
 534     fi
 535     echo "Execution failed with code ${status}."
 536     KEEP=true
 537 fi
 538
 539 smpirun_cleanup
 540
 541 exit $status