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