src/smpi/smpirun.in

   1 #! /bin/sh
   2
   3 # Copyright (c) 2007-2014. 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 DEFAULT_LOOPBACK_BANDWIDTH="498000000"
  12 DEFAULT_LOOPBACK_LATENCY="0.000004"
  13 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
  14 DEFAULT_NETWORK_LATENCY="0.000005"
  15 DEFAULT_NUMPROCS="4"
  16 DEFAULT_POWER="100"
  17
  18 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
  19 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
  20 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
  21 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
  22 POWER="${DEFAULT_POWER}"
  23
  24 SIMOPTS="--cfg=maxmin/precision:1e-3 --cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
  25
  26 #usage to print the way this script should be called
  27 usage () {
  28     cat <<EOF
  29 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
  30 Options:
  31   -keep-temps                # don't remove the generated files after execution
  32   -wrapper <command>         # use command to run the program (e.g. "valgrind")
  33   -map                       # display the machine on which each process rank is mapped
  34   -np <numprocs>             # use that amount of processes from the hostfile.
  35                              # By default, all processes of the hostfile are used.
  36   -trace-ti                  # activate time independant tracing (for replay, default in smpi_simgrid.txt)
  37   -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
  38   -trace-comment <comment>   # put a comment on the top of the trace file
  39   -trace-comment-file <file> # put file contents on the top of the trace file as comment
  40   -trace-grouped             # group MPI processes by location
  41   -trace-resource            # trace resource utilization
  42   -trace-viva                # generate configuration for Viva's GraphView
  43   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  44   -ext <value>               # additional parameter (reserved)
  45
  46 or (deprecated usage):
  47   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  48
  49 EOF
  50 }
  51
  52 #check if we have at least one parameter
  53 if [ $# -eq 0 ]
  54 then
  55     usage
  56     exit
  57 fi
  58
  59 EXTOPT=""
  60 WRAPPER=""
  61 HOSTFILE=""
  62
  63 while true; do
  64     case "$1" in
  65         "-np" | "-n")
  66             NUMPROCS="$2"
  67             shift 2
  68             ;;
  69         "-bandwidth")
  70             NETWORK_BANDWIDTH="$2"
  71             shift 2
  72             ;;
  73         "-latency")
  74             NETWORK_LATENCY="$2"
  75             shift 2
  76             ;;
  77         "-platform")
  78             PLATFORM="$2"
  79             if [ ! -f "${PLATFORM}" ]; then
  80                 echo "[$0] ** error: the file '${PLATFORM}' does not exist. Aborting."
  81                 exit 1
  82             fi
  83             shift 2
  84             ;;
  85         "-hostfile")
  86             HOSTFILE="$2"
  87             if [ ! -f "${HOSTFILE}" ]; then
  88                 echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
  89                 exit 1
  90             fi
  91             shift 2
  92             ;;
  93
  94         "-machinefile")
  95             HOSTFILE="$2"
  96             if [ ! -f "${HOSTFILE}" ]; then
  97                 echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
  98                 exit 1
  99             fi
 100             shift 2
 101             ;;
 102
 103         "-ext")
 104             EXTOPT="$2"
 105             shift 2
 106             ;;
 107
 108         "-map")
 109             MAPOPT="true"
 110             shift 1
 111             ;;
 112
 113         "-trace")
 114             TRACE_ACTIVE="true"
 115             shift 1
 116             ;;
 117
 118         "-trace-ti")
 119             TRACE_ACTIVE="true"
 120             TRACE_TI_ACTIVE="true"
 121             shift 1
 122             ;;
 123
 124         "-trace-comment")
 125             TRACE_COMMENT="$2"
 126             shift 2
 127             ;;
 128
 129         "-trace-comment-file")
 130             TRACE_COMMENT_FILE="$2"
 131             shift 2
 132             ;;
 133
 134         "-trace-file")
 135             TRACE_FILENAME="$2"
 136             shift 2
 137             ;;
 138
 139         "-trace-grouped")
 140             TRACE_GROUPED="true"
 141             shift 1
 142             ;;
 143
 144         "-trace-resource")
 145             TRACE_RESOURCE="true"
 146             shift 1
 147             ;;
 148
 149         "-trace-viva")
 150             TRACE_VIVA="true"
 151             shift 1
 152             ;;
 153
 154         "-keep-temps")
 155             KEEP="true"
 156             shift 1
 157             ;;
 158
 159         "-wrapper")
 160             WRAPPER="$2"
 161             shift 2
 162             ;;
 163
 164         "-help" | "--help" | "-h")
 165             usage
 166             exit
 167             ;;
 168
 169         "--cfg="*|"--log="*)
 170             for OPT in ${1#*=}
 171             do
 172                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 173             done
 174             shift 1
 175             ;;
 176         *)
 177             break
 178             ;;
 179     esac
 180 done
 181
 182 if [ -n "$WRAPPER" ]; then
 183     EXEC="$WRAPPER $1"
 184 else
 185     EXEC="$1"
 186 fi
 187 shift
 188
 189 # steel --cfg and --logs options
 190 while [ $# -gt 0 ]; do
 191     case "$1" in
 192         "--cfg="*|"--log="*)
 193             for OPT in ${1#*=}
 194             do
 195                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 196             done
 197             shift 1
 198             ;;
 199         *)
 200             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 201             shift
 202             ;;
 203     esac
 204 done
 205
 206
 207 ##-----------------------------------
 208
 209
 210 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 211     echo "No hostfile nor platform specified."
 212     usage
 213     exit 1
 214 fi
 215
 216 HOSTFILETMP=0
 217 if [ -z "${HOSTFILE}" ] ; then
 218     HOSTFILETMP=1
 219     HOSTFILE="$(mktemp tmphostXXXXXX)"
 220     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 221 fi
 222 UNROLLEDHOSTFILETMP=0
 223
 224 #parse if our lines are terminated by :num_process
 225 multiple_processes=`grep -c ":" $HOSTFILE`
 226 if [ "${multiple_processes}" -gt 0 ] ; then
 227     UNROLLEDHOSTFILETMP=1
 228     UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)"
 229     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 230     if [ ${HOSTFILETMP} = 1 ] ; then
 231         rm ${HOSTFILE}
 232         HOSTFILETMP=0
 233     fi
 234     HOSTFILE=$UNROLLEDHOSTFILE
 235 fi
 236
 237
 238 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 239 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 240
 241 if [ -z "${NUMPROCS}" ] ; then
 242     # Use the amount of processes in the hostfile as default value for the -np parameter
 243     NUMPROCS=$hostfile_procs
 244 fi
 245
 246
 247 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 248     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 249 fi
 250
 251 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 252 if [ -z "${PLATFORM}" ]; then
 253     PLATFORMTMP="$(mktemp tmpXXXXXX)"
 254
 255     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 256 <?xml version='1.0'?>
 257 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 258 <platform version="3">
 259 <AS id="AS0" routing="Full">
 260 PLATFORMHEAD
 261
 262     i=${NUMPROCS}
 263     while [ $i -gt 0 ]; do
 264         echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
 265         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 266         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 267         i=$((i - 1))
 268     done
 269
 270     i=${NUMPROCS}
 271     while [ $i -gt 0 ]; do
 272         j=${NUMPROCS}
 273         while [ $j -gt 0 ]; do
 274             if [ $i -eq $j ]; then
 275                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 276             else
 277                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 278             fi
 279             j=$((j - 1))
 280         done
 281         i=$((i - 1))
 282     done
 283
 284     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 285 </AS>
 286 </platform>
 287 PLATFORMFOOT
 288
 289 else
 290     PLATFORMTMP=${PLATFORM}
 291 fi
 292 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 293
 294 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 295 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
 296 #APPLICATIONTMP="app.xml"
 297
 298
 299 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 300 <?xml version='1.0'?>
 301 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 302 <platform version="3">
 303 APPLICATIONHEAD
 304
 305 ##---- cache hostnames of hostfile---------------
 306 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 307     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 308     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
 309 fi
 310
 311 if [ "${EXTOPT}" = "smpi_replay" ]; then
 312     APP_TRACES=$PROC_ARGS
 313     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 314         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 315         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 316     else
 317         printf "File not found: %s\n", "${APP_TRACES:-\${APP_TRACES\}}" >&2
 318         exit 1
 319     fi
 320 fi
 321
 322 ##----------------------------------------------------------
 323 ##  generate application.xml with hostnames from hostfile:
 324 ##  the name of host_i (1<=i<=p, where -np p) is the line i
 325 ##  in hostfile (where -hostfile hostfile), or "host$i" if
 326 ##  hostfile has less than i lines.
 327 ##----------------------------------------------------------
 328
 329 HAVE_SEQ="`which seq 2>/dev/null`"
 330
 331 if [ -n "${HAVE_SEQ}" ]; then
 332     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 333 else
 334     cnt=0
 335     while (( $cnt < ${NUMPROCS} )) ; do
 336         SEQ="$SEQ $cnt"
 337         cnt=$((cnt + 1));
 338     done
 339 fi
 340
 341 ##---- generate <process> tags------------------------------
 342
 343 for i in ${SEQ}
 344 do
 345     if [ -n "${HOSTFILE}" ]; then
 346         j=$(( $i % ${NUMHOSTS} + 1 ))
 347     fi
 348     ##---- optional display of ranks to process mapping
 349     if [ -n "${MAPOPT}" ]; then
 350         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
 351     fi
 352
 353     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
 354         host="host"$($j)
 355     else
 356         host="$(echo $hostnames|cut -d' ' -f$j)"
 357     fi
 358     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
 359     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
 360     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 361     if [ "${EXTOPT}" = "smpi_replay" ]; then
 362         if  [ ${NUMTRACES} -gt 1 ]; then
 363             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 364         else
 365             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 366         fi
 367     else
 368         for ARG in $PROC_ARGS; do
 369             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
 370         done
 371     fi
 372     echo "  </process>" >> ${APPLICATIONTMP}
 373 done
 374
 375 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 376 </platform>
 377 APPLICATIONFOOT
 378 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 379
 380 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 381 if [ -n "${TRACE_ACTIVE}" ]; then
 382     #define trace filename
 383     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 384         if [ -z "${TRACE_FILENAME}" ]; then
 385             TRACE_FILENAME="smpi_simgrid.txt"
 386         fi
 387         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 388     else
 389         if [ -z "${TRACE_FILENAME}" ]; then
 390             TRACE_FILENAME="smpi_simgrid.trace"
 391         fi
 392         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 393     fi
 394
 395     if [ -n "${TRACE_COMMENT}" ]; then
 396         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 397     fi
 398
 399     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 400         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
 401     fi
 402
 403     if [ -n "${TRACE_GROUPED}" ]; then
 404         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 405     fi
 406
 407     if [ -n "${TRACE_RESOURCE}" ]; then
 408         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 409     fi
 410
 411     if [ -n "${TRACE_VIVA}" ]; then
 412         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 413     fi
 414 fi
 415 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 416
 417 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 418 if [ -n "${KEEP}" ] ; then
 419     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 420     if [ ${HOSTFILETMP} = 1 ] ; then
 421         echo "Generated hostfile ${HOSTFILE} keeped."
 422     fi
 423     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 424         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped."
 425     fi
 426 fi
 427 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 428 status=$?
 429
 430 if [ -z "${KEEP}" ] ; then
 431     if [ -z "${PLATFORM}" ]; then
 432         rm ${PLATFORMTMP}
 433     fi
 434     if [ ${HOSTFILETMP} = 1 ] ; then
 435         rm ${HOSTFILE}
 436     fi
 437     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 438         rm ${UNROLLEDHOSTFILE}
 439     fi
 440     rm ${APPLICATIONTMP}
 441 fi
 442
 443 exit $status