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")
 151             HOSTFILE="$2"
 152             if [ ! -f "${HOSTFILE}" ]; then
 153                 die "the file '${HOSTFILE}' does not exist"
 154             fi
 155             shift 2
 156             ;;
 157         "-machinefile")
 158             HOSTFILE="$2"
 159             if [ ! -f "${HOSTFILE}" ]; then
 160                 die "the file '${HOSTFILE}' does not exist"
 161             fi
 162             shift 2
 163             ;;
 164         "-ext")
 165             EXTOPT="$2"
 166             shift 2
 167             ;;
 168         "-no-privatize")
 169             PRIVATIZE=""
 170             shift 1
 171             ;;
 172         "-map")
 173             MAPOPT=1
 174             shift 1
 175             ;;
 176         "-trace")
 177             TRACE_ACTIVE="true"
 178             shift 1
 179             ;;
 180         "-trace-ti")
 181             TRACE_ACTIVE="true"
 182             TRACE_TI_ACTIVE="true"
 183             shift 1
 184             ;;
 185         "-trace-comment")
 186             TRACE_COMMENT="$2"
 187             shift 2
 188             ;;
 189         "-trace-comment-file")
 190             TRACE_COMMENT_FILE="$2"
 191             shift 2
 192             ;;
 193         "-trace-file")
 194             TRACE_FILENAME="$2"
 195             shift 2
 196             ;;
 197         "-trace-grouped")
 198             TRACE_GROUPED="true"
 199             shift 1
 200             ;;
 201         "-trace-resource")
 202             TRACE_RESOURCE="true"
 203             shift 1
 204             ;;
 205         "-trace-viva")
 206             TRACE_VIVA="true"
 207             shift 1
 208             ;;
 209         "-keep-temps")
 210             KEEP="true"
 211             SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
 212             shift 1
 213             ;;
 214         "-wrapper")
 215             WRAPPER="$2"
 216             shift 2
 217             ;;
 218         "-help" | "--help" | "-h")
 219             usage
 220             exit 0
 221             ;;
 222         "-version" | "--version" | "-v")
 223             printf '%b\n' "$SIMGRID_VERSION"
 224             exit 0
 225             ;;
 226         "-git-version" | "--git-version")
 227             printf '%b\n' "$SIMGRID_GITHASH"
 228             exit 0
 229             ;;
 230         "--cfg="*|"--log="*)
 231             for OPT in ${1#*=}
 232             do
 233                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 234             done
 235             shift 1
 236             ;;
 237         "-foreground")
 238             # Nothing to do, compatibility.
 239             shift 1
 240             ;;
 241         *)
 242             break
 243             ;;
 244     esac
 245 done
 246
 247 # check if we still have at least one parameter beyond options
 248 if [ $# -eq 0 ]
 249 then
 250     echo "Error: no program to execute!"
 251     usage
 252     exit
 253 fi
 254
 255 EXEC="$1"
 256 shift
 257
 258 # steel --cfg and --logs options
 259 while [ $# -gt 0 ]; do
 260     case "$1" in
 261         "--cfg="*|"--log="*)
 262             for OPT in ${1#*=}
 263             do
 264                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 265             done
 266             shift 1
 267             ;;
 268         *)
 269             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 270             shift
 271             ;;
 272     esac
 273 done
 274
 275 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 276     echo "No hostfile nor platform specified."
 277     usage
 278     exit 1
 279 fi
 280
 281 if [ -z "${HOSTFILE}" ] ; then
 282     HOSTFILETMP=1
 283     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 284     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 285     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 286                 my ($pre,$rad,$post)=($1,$2,$3);
 287                 for my $elm (split(",",$rad)) {
 288                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 289                      for (my $i=$1; $i<=$2;$i++) {
 290                         print "$pre$i$post\n";
 291                      }
 292                   } else {
 293                      print "$pre$elm$post\n";
 294                   }
 295                 }
 296               } elsif (/<cluster/) {
 297             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)");
 298               }' ${PLATFORM} >> ${HOSTFILE}
 299 fi
 300 UNROLLEDHOSTFILETMP=0
 301
 302 #parse if our lines are terminated by :num_process
 303 multiple_processes=$(grep -c ":" $HOSTFILE)
 304 if [ "${multiple_processes}" -gt 0 ] ; then
 305     UNROLLEDHOSTFILETMP=1
 306     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 307     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 308     if [ ${HOSTFILETMP} = 1 ] ; then
 309         rm ${HOSTFILE}
 310         HOSTFILETMP=0
 311     fi
 312     HOSTFILE=$UNROLLEDHOSTFILE
 313 fi
 314
 315 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 316 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
 317 if [ ${hostfile_procs} = 0 ] ; then
 318     die "the hostfile '${HOSTFILE}' is empty"
 319 fi
 320
 321 if [ -z "${NUMPROCS}" ] ; then
 322     # Use the amount of processes in the hostfile as default value for the -np parameter
 323     NUMPROCS=$hostfile_procs
 324 fi
 325
 326 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 327     echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 328 fi
 329
 330 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 331 if [ -z "${PLATFORM}" ]; then
 332     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 333
 334     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 335 <?xml version='1.0'?>
 336 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 337 <platform version="4.1">
 338 <zone id="AS0" routing="Full">
 339 PLATFORMHEAD
 340
 341     i=${NUMPROCS}
 342     while [ $i -gt 0 ]; do
 343         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 344         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 345         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 346         i=$((i - 1))
 347     done
 348
 349     i=${NUMPROCS}
 350     while [ $i -gt 0 ]; do
 351         j=${NUMPROCS}
 352         while [ $j -gt 0 ]; do
 353             if [ $i -eq $j ]; then
 354                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 355             else
 356                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 357             fi
 358             j=$((j - 1))
 359         done
 360         i=$((i - 1))
 361     done
 362
 363     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 364 </zone>
 365 </platform>
 366 PLATFORMFOOT
 367
 368 else
 369     PLATFORMTMP=${PLATFORM}
 370 fi
 371 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 372 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 373 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 374 #APPLICATIONTMP="app.xml"
 375
 376 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 377 <?xml version='1.0'?>
 378 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 379 <platform version="4.1">
 380 APPLICATIONHEAD
 381
 382 ##---- cache hostnames of hostfile---------------
 383 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 384     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 385 fi
 386
 387 if [ "${EXTOPT}" = "smpi_replay" ]; then
 388     APP_TRACES=$PROC_ARGS
 389     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 390         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 391         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 392         REPLAY=1
 393     else
 394         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 395         exit 1
 396     fi
 397 fi
 398
 399 ##----------------------------------------------------------
 400 ##  generate application.xml with hostnames from hostfile:
 401 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 402 ##  hostfile has less than i lines.
 403 ##----------------------------------------------------------
 404
 405 HAVE_SEQ="$(which seq 2>/dev/null)"
 406
 407 if [ -n "${HAVE_SEQ}" ]; then
 408     SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
 409 else
 410     cnt=0
 411     while [ $cnt -lt ${NUMPROCS} ] ; do
 412         SEQ="$SEQ $cnt"
 413         cnt=$((cnt + 1));
 414     done
 415 fi
 416
 417 set -- $hostnames
 418
 419 ##---- generate <actor> tags------------------------------
 420 #prepare arguments at once
 421 for ARG in $PROC_ARGS; do
 422   XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
 423 "
 424 done
 425
 426 for i in ${SEQ}
 427 do
 428     j=$(( i % hostfile_procs + 1 ))
 429     host=$(eval "echo \${$j}")
 430
 431     ##---- optional display of ranks to actor mapping
 432     if [ ${MAPOPT} = 1 ]; then
 433       echo "[rank $i] -> $host"
 434     fi
 435
 436     echo "  <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
 437     <argument value=\"smpirun\"/> <!-- instance -->
 438     <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 439     if [ ${REPLAY} = 1 ]; then
 440         if  [ ${NUMTRACES} -gt 1 ]; then
 441             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 442         else
 443             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 444         fi
 445     else
 446     echo ${XML_ARGS} >> ${APPLICATIONTMP}
 447     fi
 448     echo "  </actor>" >> ${APPLICATIONTMP}
 449 done
 450
 451 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 452 </platform>
 453 APPLICATIONFOOT
 454 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 455 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 456 if [ -n "${TRACE_ACTIVE}" ]; then
 457     #define trace filename
 458     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 459         if [ -z "${TRACE_FILENAME}" ]; then
 460             TRACE_FILENAME="smpi_simgrid.txt"
 461         fi
 462         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 463     else
 464         if [ -z "${TRACE_FILENAME}" ]; then
 465             TRACE_FILENAME="smpi_simgrid.trace"
 466         fi
 467         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 468     fi
 469
 470     if [ -n "${TRACE_COMMENT}" ]; then
 471         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 472     fi
 473
 474     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 475         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 476     fi
 477
 478     if [ -n "${TRACE_GROUPED}" ]; then
 479         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 480     fi
 481
 482     if [ -n "${TRACE_RESOURCE}" ]; then
 483         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 484     fi
 485
 486     if [ -n "${TRACE_VIVA}" ]; then
 487         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 488     fi
 489 fi
 490 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 491
 492 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 493 if [ -n "${KEEP}" ] ; then
 494     echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 495     if [ ${HOSTFILETMP} = 1 ] ; then
 496         echo "Generated hostfile ${HOSTFILE} kept."
 497     fi
 498     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 499         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 500     fi
 501 fi
 502
 503 # Execute the process
 504 #
 505 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 506 #
 507 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 508 #
 509 # * The job is launched in the background in order to be able to handle signals.
 510 #
 511 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 512 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 513 exec 3<&0
 514 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 515 pid=$!
 516 exec 3>&-
 517 wait $pid
 518 status=$?
 519 pid=""
 520
 521 # Keep temporary files on failures to help debugging
 522 #
 523 if [ ${status} -ne 0 ] ; then
 524     if [ -z "${KEEP}" ]; then
 525         echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 526         if [ ${HOSTFILETMP} = 1 ] ; then
 527             echo "Generated hostfile ${HOSTFILE} kept."
 528         fi
 529         if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 530             echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 531         fi
 532     fi
 533     echo "Execution failed with code ${status}."
 534     KEEP=true
 535 fi
 536
 537 smpirun_cleanup
 538
 539 exit $status