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