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
 151         "-machinefile")
 152             HOSTFILE="$2"
 153             if [ ! -f "${HOSTFILE}" ]; then
 154                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 155                 exit 1
 156             fi
 157             shift 2
 158             ;;
 159
 160         "-ext")
 161             EXTOPT="$2"
 162             shift 2
 163             ;;
 164
 165         "-map")
 166             MAPOPT="true"
 167             shift 1
 168             ;;
 169
 170         "-trace")
 171             TRACE_ACTIVE="true"
 172             shift 1
 173             ;;
 174
 175         "-trace-ti")
 176             TRACE_ACTIVE="true"
 177             TRACE_TI_ACTIVE="true"
 178             shift 1
 179             ;;
 180
 181         "-trace-comment")
 182             TRACE_COMMENT="$2"
 183             shift 2
 184             ;;
 185
 186         "-trace-comment-file")
 187             TRACE_COMMENT_FILE="$2"
 188             shift 2
 189             ;;
 190
 191         "-trace-file")
 192             TRACE_FILENAME="$2"
 193             shift 2
 194             ;;
 195
 196         "-trace-grouped")
 197             TRACE_GROUPED="true"
 198             shift 1
 199             ;;
 200
 201         "-trace-resource")
 202             TRACE_RESOURCE="true"
 203             shift 1
 204             ;;
 205
 206         "-trace-viva")
 207             TRACE_VIVA="true"
 208             shift 1
 209             ;;
 210
 211         "-keep-temps")
 212             KEEP="true"
 213             shift 1
 214             ;;
 215
 216         "-wrapper")
 217             WRAPPER="$2"
 218             shift 2
 219             ;;
 220
 221         "-help" | "--help" | "-h")
 222             usage
 223             exit 0
 224             ;;
 225
 226         "-version" | "--version" | "-v")
 227             printf '%b\n' "$SIMGRID_VERSION"
 228             exit 0
 229             ;;
 230
 231         "-git-version" | "--git-version")
 232             printf '%b\n' "$SIMGRID_GITHASH"
 233             exit 0
 234             ;;
 235
 236         "--cfg="*|"--log="*)
 237             for OPT in ${1#*=}
 238             do
 239                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 240             done
 241             shift 1
 242             ;;
 243         "-foreground")
 244             # Nothing to do, compatibility.
 245             shift 1
 246             ;;
 247         *)
 248             break
 249             ;;
 250     esac
 251 done
 252
 253 if [ -n "$WRAPPER" ]; then
 254     EXEC="$WRAPPER $1"
 255 else
 256     EXEC="$1"
 257 fi
 258 shift
 259
 260 # steel --cfg and --logs options
 261 while [ $# -gt 0 ]; do
 262     case "$1" in
 263         "--cfg="*|"--log="*)
 264             for OPT in ${1#*=}
 265             do
 266                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 267             done
 268             shift 1
 269             ;;
 270         *)
 271             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 272             shift
 273             ;;
 274     esac
 275 done
 276
 277
 278 ##-----------------------------------
 279
 280
 281 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 282     echo "No hostfile nor platform specified."
 283     usage
 284     exit 1
 285 fi
 286
 287 if [ -z "${HOSTFILE}" ] ; then
 288     HOSTFILETMP=1
 289     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 290     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 291     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 292                 my ($pre,$rad,$post)=($1,$2,$3);
 293                 for my $elm (split(",",$rad)) {
 294                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 295                      for (my $i=$1; $i<=$2;$i++) {
 296                         print "$pre$i$post\n";
 297                      }
 298                   } else {
 299                      print "$pre$elm$post\n";
 300                   }
 301                 }
 302               } elsif (/<cluster/) {
 303                 die ("Unparsable cluster tag. Either provide an hostfile yourself or give the attributes prefix, radical and suffix in that order on the <cluster line");
 304               }' ${PLATFORM} >> ${HOSTFILE}
 305 fi
 306 UNROLLEDHOSTFILETMP=0
 307
 308 #parse if our lines are terminated by :num_process
 309 multiple_processes=`grep -c ":" $HOSTFILE`
 310 if [ "${multiple_processes}" -gt 0 ] ; then
 311     UNROLLEDHOSTFILETMP=1
 312     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 313     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 314     if [ ${HOSTFILETMP} = 1 ] ; then
 315         rm ${HOSTFILE}
 316         HOSTFILETMP=0
 317     fi
 318     HOSTFILE=$UNROLLEDHOSTFILE
 319 fi
 320
 321
 322 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 323 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 324 if [ ${hostfile_procs} = 0 ] ; then
 325    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
 326    exit 1
 327 fi
 328
 329 if [ -z "${NUMPROCS}" ] ; then
 330     # Use the amount of processes in the hostfile as default value for the -np parameter
 331     NUMPROCS=$hostfile_procs
 332 fi
 333
 334
 335 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 336     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 337 fi
 338
 339 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 340 if [ -z "${PLATFORM}" ]; then
 341     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 342
 343     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 344 <?xml version='1.0'?>
 345 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 346 <platform version="4">
 347 <AS id="AS0" routing="Full">
 348 PLATFORMHEAD
 349
 350     i=${NUMPROCS}
 351     while [ $i -gt 0 ]; do
 352         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 353         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 354         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 355         i=$((i - 1))
 356     done
 357
 358     i=${NUMPROCS}
 359     while [ $i -gt 0 ]; do
 360         j=${NUMPROCS}
 361         while [ $j -gt 0 ]; do
 362             if [ $i -eq $j ]; then
 363                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 364             else
 365                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 366             fi
 367             j=$((j - 1))
 368         done
 369         i=$((i - 1))
 370     done
 371
 372     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 373 </AS>
 374 </platform>
 375 PLATFORMFOOT
 376
 377 else
 378     PLATFORMTMP=${PLATFORM}
 379 fi
 380 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 381
 382 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 383 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 384 #APPLICATIONTMP="app.xml"
 385
 386
 387 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 388 <?xml version='1.0'?>
 389 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 390 <platform version="4">
 391 APPLICATIONHEAD
 392
 393 ##---- cache hostnames of hostfile---------------
 394 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 395     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 396     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
 397 fi
 398
 399 if [ "${EXTOPT}" = "smpi_replay" ]; then
 400     APP_TRACES=$PROC_ARGS
 401     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 402         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 403         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 404     else
 405         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 406         exit 1
 407     fi
 408 fi
 409
 410 ##----------------------------------------------------------
 411 ##  generate application.xml with hostnames from hostfile:
 412 ##  the name of host_i (1<=i<=p, where -np p) is the line i
 413 ##  in hostfile (where -hostfile hostfile), or "host$i" if
 414 ##  hostfile has less than i lines.
 415 ##----------------------------------------------------------
 416
 417 HAVE_SEQ="`which seq 2>/dev/null`"
 418
 419 if [ -n "${HAVE_SEQ}" ]; then
 420     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 421 else
 422     cnt=0
 423     while [ $cnt -lt ${NUMPROCS} ] ; do
 424         SEQ="$SEQ $cnt"
 425         cnt=$((cnt + 1));
 426     done
 427 fi
 428
 429 ##---- generate <process> tags------------------------------
 430
 431 for i in ${SEQ}
 432 do
 433     if [ -n "${HOSTFILE}" ]; then
 434         j=$(( $i % ${NUMHOSTS} + 1 ))
 435     fi
 436     ##---- optional display of ranks to process mapping
 437     if [ -n "${MAPOPT}" ]; then
 438         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
 439     fi
 440
 441     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
 442         host="host"$($j)
 443     else
 444         host="$(echo $hostnames|cut -d' ' -f$j)"
 445     fi
 446     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
 447     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
 448     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 449     if [ "${EXTOPT}" = "smpi_replay" ]; then
 450         if  [ ${NUMTRACES} -gt 1 ]; then
 451             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 452         else
 453             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 454         fi
 455     else
 456         for ARG in $PROC_ARGS; do
 457             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
 458         done
 459     fi
 460     echo "  </process>" >> ${APPLICATIONTMP}
 461 done
 462
 463 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 464 </platform>
 465 APPLICATIONFOOT
 466 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 467
 468 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 469 if [ -n "${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"
 474         fi
 475         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 476     else
 477         if [ -z "${TRACE_FILENAME}" ]; then
 478             TRACE_FILENAME="smpi_simgrid.trace"
 479         fi
 480         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 481     fi
 482
 483     if [ -n "${TRACE_COMMENT}" ]; then
 484         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 485     fi
 486
 487     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 488         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
 489     fi
 490
 491     if [ -n "${TRACE_GROUPED}" ]; then
 492         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 493     fi
 494
 495     if [ -n "${TRACE_RESOURCE}" ]; then
 496         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 497     fi
 498
 499     if [ -n "${TRACE_VIVA}" ]; then
 500         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 501     fi
 502 fi
 503 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 504
 505 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 506 if [ -n "${KEEP}" ] ; then
 507     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 508     if [ ${HOSTFILETMP} = 1 ] ; then
 509         echo "Generated hostfile ${HOSTFILE} kept."
 510     fi
 511     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 512         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 513     fi
 514 fi
 515
 516 # Execute the process
 517 #
 518 # The shell still need to be alive for the duration in order to do some cleanup
 519 # after the process.
 520 #
 521 # We are going through great lengths in order to both keep stdin and be able
 522 # to handle signals:
 523 #
 524 # * The job is launched in the background in order to be able to handle
 525 #   signals.
 526 #
 527 # * The FD 3 is used to temporarily store FD 1. This is because the shell
 528 #   connects FD 1 to /dev/null when the command is launched in the
 529 #   background: this can be overriden in bash but not in standard bourne shell.
 530 exec 3<&0
 531 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 532 pid=$!
 533 exec 3>&-
 534 wait $pid
 535 status=$?
 536 pid=""
 537
 538 smpirun_cleanup
 539
 540 exit $status