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 SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
  12
  13 DEFAULT_LOOPBACK_BANDWIDTH="498000000"
  14 DEFAULT_LOOPBACK_LATENCY="0.000004"
  15 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
  16 DEFAULT_NETWORK_LATENCY="0.000005"
  17 DEFAULT_NUMPROCS="4"
  18 DEFAULT_POWER="100"
  19
  20 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
  21 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
  22 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
  23 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
  24 POWER="${DEFAULT_POWER}"
  25
  26 SIMOPTS="--cfg=maxmin/precision:1e-3 --cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
  27
  28 #usage to print the way this script should be called
  29 usage () {
  30     cat <<EOF
  31 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
  32 Options:
  33   -keep-temps                # don't remove the generated files after execution
  34   -wrapper <command>         # use command to run the program (e.g. "valgrind")
  35   -map                       # display the machine on which each process rank is mapped
  36   -np <numprocs>             # use that amount of processes from the hostfile.
  37                              # By default, all processes of the hostfile are used.
  38   -trace-ti                  # activate time independant tracing (for replay, default in smpi_simgrid.txt)
  39   -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
  40   -trace-comment <comment>   # put a comment on the top of the trace file
  41   -trace-comment-file <file> # put file contents on the top of the trace file as comment
  42   -trace-grouped             # group MPI processes by location
  43   -trace-resource            # trace resource utilization
  44   -trace-viva                # generate configuration for Viva's GraphView
  45   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  46   -ext <value>               # additional parameter (reserved)
  47
  48 or (deprecated usage):
  49   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  50
  51 EOF
  52 }
  53
  54 #check if we have at least one parameter
  55 if [ $# -eq 0 ]
  56 then
  57     usage
  58     exit
  59 fi
  60
  61 EXTOPT=""
  62 WRAPPER=""
  63 HOSTFILE=""
  64
  65 while true; do
  66     case "$1" in
  67         "-np" | "-n")
  68             NUMPROCS="$2"
  69             shift 2
  70             ;;
  71         "-bandwidth")
  72             NETWORK_BANDWIDTH="$2"
  73             shift 2
  74             ;;
  75         "-latency")
  76             NETWORK_LATENCY="$2"
  77             shift 2
  78             ;;
  79         "-platform")
  80             PLATFORM="$2"
  81             if [ ! -f "${PLATFORM}" ]; then
  82                 echo "[$0] ** error: the file '${PLATFORM}' does not exist. Aborting."
  83                 exit 1
  84             fi
  85             shift 2
  86             ;;
  87         "-hostfile")
  88             HOSTFILE="$2"
  89             if [ ! -f "${HOSTFILE}" ]; then
  90                 echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
  91                 exit 1
  92             fi
  93             shift 2
  94             ;;
  95
  96         "-machinefile")
  97             HOSTFILE="$2"
  98             if [ ! -f "${HOSTFILE}" ]; then
  99                 echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
 100                 exit 1
 101             fi
 102             shift 2
 103             ;;
 104
 105         "-ext")
 106             EXTOPT="$2"
 107             shift 2
 108             ;;
 109
 110         "-map")
 111             MAPOPT="true"
 112             shift 1
 113             ;;
 114
 115         "-trace")
 116             TRACE_ACTIVE="true"
 117             shift 1
 118             ;;
 119
 120         "-trace-ti")
 121             TRACE_ACTIVE="true"
 122             TRACE_TI_ACTIVE="true"
 123             shift 1
 124             ;;
 125
 126         "-trace-comment")
 127             TRACE_COMMENT="$2"
 128             shift 2
 129             ;;
 130
 131         "-trace-comment-file")
 132             TRACE_COMMENT_FILE="$2"
 133             shift 2
 134             ;;
 135
 136         "-trace-file")
 137             TRACE_FILENAME="$2"
 138             shift 2
 139             ;;
 140
 141         "-trace-grouped")
 142             TRACE_GROUPED="true"
 143             shift 1
 144             ;;
 145
 146         "-trace-resource")
 147             TRACE_RESOURCE="true"
 148             shift 1
 149             ;;
 150
 151         "-trace-viva")
 152             TRACE_VIVA="true"
 153             shift 1
 154             ;;
 155
 156         "-keep-temps")
 157             KEEP="true"
 158             shift 1
 159             ;;
 160
 161         "-wrapper")
 162             WRAPPER="$2"
 163             shift 2
 164             ;;
 165
 166         "-help" | "--help" | "-h")
 167             usage
 168             exit 0
 169             ;;
 170
 171         "-version" | "--version" | "-v")
 172             printf '%b\n' "$SIMGRID_VERSION"
 173             exit 0
 174             ;;
 175
 176         "--cfg="*|"--log="*)
 177             for OPT in ${1#*=}
 178             do
 179                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 180             done
 181             shift 1
 182             ;;
 183         *)
 184             break
 185             ;;
 186     esac
 187 done
 188
 189 if [ -n "$WRAPPER" ]; then
 190     EXEC="$WRAPPER $1"
 191 else
 192     EXEC="$1"
 193 fi
 194 shift
 195
 196 # steel --cfg and --logs options
 197 while [ $# -gt 0 ]; do
 198     case "$1" in
 199         "--cfg="*|"--log="*)
 200             for OPT in ${1#*=}
 201             do
 202                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 203             done
 204             shift 1
 205             ;;
 206         *)
 207             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 208             shift
 209             ;;
 210     esac
 211 done
 212
 213
 214 ##-----------------------------------
 215
 216
 217 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 218     echo "No hostfile nor platform specified."
 219     usage
 220     exit 1
 221 fi
 222
 223 HOSTFILETMP=0
 224 if [ -z "${HOSTFILE}" ] ; then
 225     HOSTFILETMP=1
 226     HOSTFILE="$(mktemp tmphostXXXXXX)"
 227     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 228 fi
 229 UNROLLEDHOSTFILETMP=0
 230
 231 #parse if our lines are terminated by :num_process
 232 multiple_processes=`grep -c ":" $HOSTFILE`
 233 if [ "${multiple_processes}" -gt 0 ] ; then
 234     UNROLLEDHOSTFILETMP=1
 235     UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)"
 236     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 237     if [ ${HOSTFILETMP} = 1 ] ; then
 238         rm ${HOSTFILE}
 239         HOSTFILETMP=0
 240     fi
 241     HOSTFILE=$UNROLLEDHOSTFILE
 242 fi
 243
 244
 245 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 246 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
 247
 248 if [ -z "${NUMPROCS}" ] ; then
 249     # Use the amount of processes in the hostfile as default value for the -np parameter
 250     NUMPROCS=$hostfile_procs
 251 fi
 252
 253
 254 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 255     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 256 fi
 257
 258 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 259 if [ -z "${PLATFORM}" ]; then
 260     PLATFORMTMP="$(mktemp tmpXXXXXX)"
 261
 262     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 263 <?xml version='1.0'?>
 264 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 265 <platform version="3">
 266 <AS id="AS0" routing="Full">
 267 PLATFORMHEAD
 268
 269     i=${NUMPROCS}
 270     while [ $i -gt 0 ]; do
 271         echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
 272         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 273         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 274         i=$((i - 1))
 275     done
 276
 277     i=${NUMPROCS}
 278     while [ $i -gt 0 ]; do
 279         j=${NUMPROCS}
 280         while [ $j -gt 0 ]; do
 281             if [ $i -eq $j ]; then
 282                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 283             else
 284                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 285             fi
 286             j=$((j - 1))
 287         done
 288         i=$((i - 1))
 289     done
 290
 291     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 292 </AS>
 293 </platform>
 294 PLATFORMFOOT
 295
 296 else
 297     PLATFORMTMP=${PLATFORM}
 298 fi
 299 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 300
 301 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 302 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
 303 #APPLICATIONTMP="app.xml"
 304
 305
 306 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 307 <?xml version='1.0'?>
 308 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 309 <platform version="3">
 310 APPLICATIONHEAD
 311
 312 ##---- cache hostnames of hostfile---------------
 313 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 314     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 315     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
 316 fi
 317
 318 if [ "${EXTOPT}" = "smpi_replay" ]; then
 319     APP_TRACES=$PROC_ARGS
 320     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 321         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 322         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 323     else
 324         printf "File not found: %s\n", "${APP_TRACES:-\${APP_TRACES\}}" >&2
 325         exit 1
 326     fi
 327 fi
 328
 329 ##----------------------------------------------------------
 330 ##  generate application.xml with hostnames from hostfile:
 331 ##  the name of host_i (1<=i<=p, where -np p) is the line i
 332 ##  in hostfile (where -hostfile hostfile), or "host$i" if
 333 ##  hostfile has less than i lines.
 334 ##----------------------------------------------------------
 335
 336 HAVE_SEQ="`which seq 2>/dev/null`"
 337
 338 if [ -n "${HAVE_SEQ}" ]; then
 339     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 340 else
 341     cnt=0
 342     while (( $cnt < ${NUMPROCS} )) ; do
 343         SEQ="$SEQ $cnt"
 344         cnt=$((cnt + 1));
 345     done
 346 fi
 347
 348 ##---- generate <process> tags------------------------------
 349
 350 for i in ${SEQ}
 351 do
 352     if [ -n "${HOSTFILE}" ]; then
 353         j=$(( $i % ${NUMHOSTS} + 1 ))
 354     fi
 355     ##---- optional display of ranks to process mapping
 356     if [ -n "${MAPOPT}" ]; then
 357         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
 358     fi
 359
 360     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
 361         host="host"$($j)
 362     else
 363         host="$(echo $hostnames|cut -d' ' -f$j)"
 364     fi
 365     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
 366     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
 367     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 368     if [ "${EXTOPT}" = "smpi_replay" ]; then
 369         if  [ ${NUMTRACES} -gt 1 ]; then
 370             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 371         else
 372             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 373         fi
 374     else
 375         for ARG in $PROC_ARGS; do
 376             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
 377         done
 378     fi
 379     echo "  </process>" >> ${APPLICATIONTMP}
 380 done
 381
 382 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 383 </platform>
 384 APPLICATIONFOOT
 385 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 386
 387 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 388 if [ -n "${TRACE_ACTIVE}" ]; then
 389     #define trace filename
 390     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 391         if [ -z "${TRACE_FILENAME}" ]; then
 392             TRACE_FILENAME="smpi_simgrid.txt"
 393         fi
 394         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 395     else
 396         if [ -z "${TRACE_FILENAME}" ]; then
 397             TRACE_FILENAME="smpi_simgrid.trace"
 398         fi
 399         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 400     fi
 401
 402     if [ -n "${TRACE_COMMENT}" ]; then
 403         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 404     fi
 405
 406     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 407         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
 408     fi
 409
 410     if [ -n "${TRACE_GROUPED}" ]; then
 411         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 412     fi
 413
 414     if [ -n "${TRACE_RESOURCE}" ]; then
 415         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 416     fi
 417
 418     if [ -n "${TRACE_VIVA}" ]; then
 419         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
 420     fi
 421 fi
 422 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 423
 424 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 425 if [ -n "${KEEP}" ] ; then
 426     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 427     if [ ${HOSTFILETMP} = 1 ] ; then
 428         echo "Generated hostfile ${HOSTFILE} keeped."
 429     fi
 430     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 431         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped."
 432     fi
 433 fi
 434 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 435 status=$?
 436
 437 if [ -z "${KEEP}" ] ; then
 438     if [ -z "${PLATFORM}" ]; then
 439         rm ${PLATFORMTMP}
 440     fi
 441     if [ ${HOSTFILETMP} = 1 ] ; then
 442         rm ${HOSTFILE}
 443     fi
 444     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 445         rm ${UNROLLEDHOSTFILE}
 446     fi
 447     rm ${APPLICATIONTMP}
 448 fi
 449
 450 exit $status