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="498000000"
  15 DEFAULT_LOOPBACK_LATENCY="0.000004"
  16 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
  17 DEFAULT_NETWORK_LATENCY="0.000005"
  18 DEFAULT_NUMPROCS="4"
  19 DEFAULT_POWER="100"
  20
  21 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
  22 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
  23 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
  24 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
  25 POWER="${DEFAULT_POWER}"
  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   -foreground                # run the child process in the foreground:
  49                              # - it can access the terminal;
  50                              # - but we cannot stop it when we receive a signal.
  51                              # This is useful with -wrapper "gdb --args".
  52
  53   -version                   # Displays the SimGrid version (human readable)
  54   -git-version               # Displays the git hash of SimGrid
  55
  56 or (deprecated usage):
  57   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  58
  59 EOF
  60 }
  61
  62 #check if we have at least one parameter
  63 if [ $# -eq 0 ]
  64 then
  65     usage
  66     exit
  67 fi
  68
  69 EXTOPT=""
  70 WRAPPER=""
  71 HOSTFILE=""
  72 HOSTFILETMP=0
  73
  74 unset pid
  75 unset foreground ; foreground=0
  76
  77 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
  78
  79 smpirun_cleanup()
  80 {
  81   if [ -z "${KEEP}" ] ; then
  82       if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
  83         rm -f ${PLATFORMTMP}
  84         PLATFORMTMP=""
  85       fi
  86       if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
  87           rm -f ${HOSTFILE}
  88           HOSTFILE=""
  89       fi
  90       if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
  91           rm -f ${UNROLLEDHOSTFILE}
  92           UNROLLEDHOSTFILE=""
  93       fi
  94       if [ -n ${APPLICATIONTMP} ]; then
  95         rm -f ${APPLICATIONTMP}
  96         APPLICATIONTMP=""
  97       fi
  98   fi
  99 }
 100
 101 smpirun_trap() {
 102   local sig
 103   sig="$1"
 104
 105   # Cleanup and kill the child process:
 106   smpirun_cleanup
 107   if ! [ -z "$pid" ]; then
 108     kill -TERM $pid
 109   fi
 110   unset pid
 111
 112   # Raise the same signal again (remove the traps first):
 113   trap - $trapped_signals
 114   kill -$sig $$
 115
 116   # This should never happen:
 117   kill -ABRT $$
 118   kill -TERM $$
 119 }
 120
 121 for s in $trapped_signals; do
 122   trap "smpirun_trap $s" $s
 123 done
 124
 125 while true; do
 126     case "$1" in
 127         "-np" | "-n")
 128             NUMPROCS="$2"
 129             shift 2
 130             ;;
 131         "-bandwidth")
 132             NETWORK_BANDWIDTH="$2"
 133             shift 2
 134             ;;
 135         "-latency")
 136             NETWORK_LATENCY="$2"
 137             shift 2
 138             ;;
 139         "-platform")
 140             PLATFORM="$2"
 141             if [ ! -f "${PLATFORM}" ]; then
 142                 echo "[`basename $0`] ** error: the file '${PLATFORM}' does not exist. Aborting."
 143                 exit 1
 144             fi
 145             shift 2
 146             ;;
 147         "-hostfile")
 148             HOSTFILE="$2"
 149             if [ ! -f "${HOSTFILE}" ]; then
 150                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 151                 exit 1
 152             fi
 153             shift 2
 154             ;;
 155
 156         "-machinefile")
 157             HOSTFILE="$2"
 158             if [ ! -f "${HOSTFILE}" ]; then
 159                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 160                 exit 1
 161             fi
 162             shift 2
 163             ;;
 164
 165         "-ext")
 166             EXTOPT="$2"
 167             shift 2
 168             ;;
 169
 170         "-map")
 171             MAPOPT="true"
 172             shift 1
 173             ;;
 174
 175         "-trace")
 176             TRACE_ACTIVE="true"
 177             shift 1
 178             ;;
 179
 180         "-trace-ti")
 181             TRACE_ACTIVE="true"
 182             TRACE_TI_ACTIVE="true"
 183             shift 1
 184             ;;
 185
 186         "-trace-comment")
 187             TRACE_COMMENT="$2"
 188             shift 2
 189             ;;
 190
 191         "-trace-comment-file")
 192             TRACE_COMMENT_FILE="$2"
 193             shift 2
 194             ;;
 195
 196         "-trace-file")
 197             TRACE_FILENAME="$2"
 198             shift 2
 199             ;;
 200
 201         "-trace-grouped")
 202             TRACE_GROUPED="true"
 203             shift 1
 204             ;;
 205
 206         "-trace-resource")
 207             TRACE_RESOURCE="true"
 208             shift 1
 209             ;;
 210
 211         "-trace-viva")
 212             TRACE_VIVA="true"
 213             shift 1
 214             ;;
 215
 216         "-keep-temps")
 217             KEEP="true"
 218             shift 1
 219             ;;
 220
 221         "-wrapper")
 222             WRAPPER="$2"
 223             shift 2
 224             ;;
 225
 226         "-help" | "--help" | "-h")
 227             usage
 228             exit 0
 229             ;;
 230
 231         "-version" | "--version" | "-v")
 232             printf '%b\n' "$SIMGRID_VERSION"
 233             exit 0
 234             ;;
 235
 236         "-git-version" | "--git-version")
 237             printf '%b\n' "$SIMGRID_GITHASH"
 238             exit 0
 239             ;;
 240
 241         "--cfg="*|"--log="*)
 242             for OPT in ${1#*=}
 243             do
 244                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 245             done
 246             shift 1
 247             ;;
 248         "-foreground")
 249             foreground=1
 250             shift 1
 251             ;;
 252         *)
 253             break
 254             ;;
 255     esac
 256 done
 257
 258 if [ -n "$WRAPPER" ]; then
 259     EXEC="$WRAPPER $1"
 260 else
 261     EXEC="$1"
 262 fi
 263 shift
 264
 265 # steel --cfg and --logs options
 266 while [ $# -gt 0 ]; do
 267     case "$1" in
 268         "--cfg="*|"--log="*)
 269             for OPT in ${1#*=}
 270             do
 271                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 272             done
 273             shift 1
 274             ;;
 275         *)
 276             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 277             shift
 278             ;;
 279     esac
 280 done
 281
 282
 283 ##-----------------------------------
 284
 285
 286 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 287     echo "No hostfile nor platform specified."
 288     usage
 289     exit 1
 290 fi
 291
 292 if [ -z "${HOSTFILE}" ] ; then
 293     HOSTFILETMP=1
 294     HOSTFILE="$(mktemp tmphostXXXXXX)"
 295     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 296     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 297                 my ($pre,$rad,$post)=($1,$2,$3);
 298                 for my $elm (split(",",$rad)) {
 299                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 300                      for (my $i=$1; $i<=$2;$i++) {
 301                         print "$pre$i$post\n";
 302                      }
 303                   } else {
 304                      print "$pre$elm$post\n";
 305                   }
 306                 }
 307               } elsif (/<cluster/) {
 308                 die ("Unparsable cluster tag. Either provide an hostfile yourself or give the attributes prefix, radical and suffix in that order on the <cluster line");
 309               }' ${PLATFORM} >> ${HOSTFILE}
 310 fi
 311 UNROLLEDHOSTFILETMP=0
 312
 313 #parse if our lines are terminated by :num_process
 314 multiple_processes=`grep -c ":" $HOSTFILE`
 315 if [ "${multiple_processes}" -gt 0 ] ; then
 316     UNROLLEDHOSTFILETMP=1
 317     UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)"
 318     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 319     if [ ${HOSTFILETMP} = 1 ] ; then
 320         rm ${HOSTFILE}
 321         HOSTFILETMP=0
 322     fi
 323     HOSTFILE=$UNROLLEDHOSTFILE
 324 fi
 325
 326
 327 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 328 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 329 if [ ${hostfile_procs} = 0 ] ; then
 330    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
 331    exit 1
 332 fi
 333
 334 if [ -z "${NUMPROCS}" ] ; then
 335     # Use the amount of processes in the hostfile as default value for the -np parameter
 336     NUMPROCS=$hostfile_procs
 337 fi
 338
 339
 340 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 341     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 342 fi
 343
 344 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 345 if [ -z "${PLATFORM}" ]; then
 346     PLATFORMTMP="$(mktemp tmpXXXXXX)"
 347
 348     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 349 <?xml version='1.0'?>
 350 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 351 <platform version="3">
 352 <AS id="AS0" routing="Full">
 353 PLATFORMHEAD
 354
 355     i=${NUMPROCS}
 356     while [ $i -gt 0 ]; do
 357         echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
 358         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 359         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 360         i=$((i - 1))
 361     done
 362
 363     i=${NUMPROCS}
 364     while [ $i -gt 0 ]; do
 365         j=${NUMPROCS}
 366         while [ $j -gt 0 ]; do
 367             if [ $i -eq $j ]; then
 368                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 369             else
 370                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 371             fi
 372             j=$((j - 1))
 373         done
 374         i=$((i - 1))
 375     done
 376
 377     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 378 </AS>
 379 </platform>
 380 PLATFORMFOOT
 381
 382 else
 383     PLATFORMTMP=${PLATFORM}
 384 fi
 385 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 386
 387 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 388 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
 389 #APPLICATIONTMP="app.xml"
 390
 391
 392 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 393 <?xml version='1.0'?>
 394 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 395 <platform version="3">
 396 APPLICATIONHEAD
 397
 398 ##---- cache hostnames of hostfile---------------
 399 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 400     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 401     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
 402 fi
 403
 404 if [ "${EXTOPT}" = "smpi_replay" ]; then
 405     APP_TRACES=$PROC_ARGS
 406     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 407         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 408         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 409     else
 410         printf "File not found: %s\n", "${APP_TRACES:-\${APP_TRACES\}}" >&2
 411         exit 1
 412     fi
 413 fi
 414
 415 ##----------------------------------------------------------
 416 ##  generate application.xml with hostnames from hostfile:
 417 ##  the name of host_i (1<=i<=p, where -np p) is the line i
 418 ##  in hostfile (where -hostfile hostfile), or "host$i" if
 419 ##  hostfile has less than i lines.
 420 ##----------------------------------------------------------
 421
 422 HAVE_SEQ="`which seq 2>/dev/null`"
 423
 424 if [ -n "${HAVE_SEQ}" ]; then
 425     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 426 else
 427     cnt=0
 428     while [ $cnt -lt ${NUMPROCS} ] ; do
 429         SEQ="$SEQ $cnt"
 430         cnt=$((cnt + 1));
 431     done
 432 fi
 433
 434 ##---- generate <process> tags------------------------------
 435
 436 for i in ${SEQ}
 437 do
 438     if [ -n "${HOSTFILE}" ]; then
 439         j=$(( $i % ${NUMHOSTS} + 1 ))
 440     fi
 441     ##---- optional display of ranks to process mapping
 442     if [ -n "${MAPOPT}" ]; then
 443         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
 444     fi
 445
 446     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
 447         host="host"$($j)
 448     else
 449         host="$(echo $hostnames|cut -d' ' -f$j)"
 450     fi
 451     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
 452     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
 453     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 454     if [ "${EXTOPT}" = "smpi_replay" ]; then
 455         if  [ ${NUMTRACES} -gt 1 ]; then
 456             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 457         else
 458             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 459         fi
 460     else
 461         for ARG in $PROC_ARGS; do
 462             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
 463         done
 464     fi
 465     echo "  </process>" >> ${APPLICATIONTMP}
 466 done
 467
 468 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 469 </platform>
 470 APPLICATIONFOOT
 471 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 472
 473 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 474 if [ -n "${TRACE_ACTIVE}" ]; then
 475     #define trace filename
 476     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 477         if [ -z "${TRACE_FILENAME}" ]; then
 478             TRACE_FILENAME="smpi_simgrid.txt"
 479         fi
 480         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 481     else
 482         if [ -z "${TRACE_FILENAME}" ]; then
 483             TRACE_FILENAME="smpi_simgrid.trace"
 484         fi
 485         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 486     fi
 487
 488     if [ -n "${TRACE_COMMENT}" ]; then
 489         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 490     fi
 491
 492     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 493         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
 494     fi
 495
 496     if [ -n "${TRACE_GROUPED}" ]; then
 497         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 498     fi
 499
 500     if [ -n "${TRACE_RESOURCE}" ]; then
 501         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 502     fi
 503
 504     if [ -n "${TRACE_VIVA}" ]; then
 505         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 506     fi
 507 fi
 508 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 509
 510 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 511 if [ -n "${KEEP}" ] ; then
 512     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 513     if [ ${HOSTFILETMP} = 1 ] ; then
 514         echo "Generated hostfile ${HOSTFILE} keeped."
 515     fi
 516     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 517         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped."
 518     fi
 519 fi
 520 if [ "$foreground" = 1 ]; then
 521   ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 522   status=$?
 523 else
 524   ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} &
 525   pid=$!
 526   wait $pid
 527   status=$?
 528   pid=""
 529 fi
 530
 531 smpirun_cleanup
 532
 533 exit $status