src/smpi/smpirun.in

   1 #!/usr/bin/env 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-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  48   -ext <value>               # additional parameter (reserved)
  49
  50   -version                   # Displays the SimGrid version (human readable)
  51   -git-version               # Displays the git hash of SimGrid
  52
  53 or (deprecated usage):
  54   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  55
  56 EOF
  57 }
  58
  59 #check if we have at least one parameter
  60 if [ $# -eq 0 ]
  61 then
  62     usage
  63     exit
  64 fi
  65
  66 EXTOPT=""
  67 WRAPPER=""
  68 HOSTFILE=""
  69 HOSTFILETMP=0
  70 MAPOPT=0
  71 REPLAY=0
  72
  73 unset pid
  74
  75 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
  76
  77 die () {
  78     printf '[%s] ** error: %s. Aborting.\n' "$(basename $0)" "$*" >&2
  79     exit 1
  80 }
  81
  82 smpirun_cleanup()
  83 {
  84   if [ -z "${KEEP}" ] ; then
  85       if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
  86         rm -f ${PLATFORMTMP}
  87         PLATFORMTMP=""
  88       fi
  89       if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
  90           rm -f ${HOSTFILE}
  91           HOSTFILE=""
  92       fi
  93       if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
  94           rm -f ${UNROLLEDHOSTFILE}
  95           UNROLLEDHOSTFILE=""
  96       fi
  97       if [ -n ${APPLICATIONTMP} ]; then
  98         rm -f ${APPLICATIONTMP}
  99         APPLICATIONTMP=""
 100       fi
 101   fi
 102 }
 103
 104 smpirun_trap() {
 105   local sig
 106   sig="$1"
 107
 108   # Cleanup and kill the child process:
 109   smpirun_cleanup
 110   if ! [ -z "$pid" ]; then
 111     kill -TERM $pid
 112   fi
 113   unset pid
 114
 115   # Raise the same signal again (remove the traps first):
 116   trap - $trapped_signals
 117   kill -$sig $$
 118
 119   # This should never happen:
 120   kill -ABRT $$
 121   kill -TERM $$
 122 }
 123
 124 for s in $trapped_signals; do
 125   trap "smpirun_trap $s" $s
 126 done
 127
 128 while true; do
 129     case "$1" in
 130         "-np" | "-n")
 131             NUMPROCS="$2"
 132             shift 2
 133             ;;
 134         "-bandwidth")
 135             NETWORK_BANDWIDTH="$2"
 136             shift 2
 137             ;;
 138         "-latency")
 139             NETWORK_LATENCY="$2"
 140             shift 2
 141             ;;
 142         "-platform")
 143             PLATFORM="$2"
 144             if [ ! -f "${PLATFORM}" ]; then
 145                 die "the file '${PLATFORM}' does not exist"
 146             fi
 147             shift 2
 148             ;;
 149         "-hostfile" | "-machinefile")
 150             HOSTFILE="$2"
 151             if [ ! -f "${HOSTFILE}" ]; then
 152                 die "the file '${HOSTFILE}' does not exist"
 153             fi
 154             shift 2
 155             ;;
 156         "-ext")
 157             EXTOPT="$2"
 158             shift 2
 159             ;;
 160         "-no-privatize")
 161             PRIVATIZE=""
 162             shift 1
 163             ;;
 164         "-map")
 165             MAPOPT=1
 166             shift 1
 167             ;;
 168         "-trace")
 169             TRACE_ACTIVE="true"
 170             shift 1
 171             ;;
 172         "-trace-ti")
 173             TRACE_ACTIVE="true"
 174             TRACE_TI_ACTIVE="true"
 175             shift 1
 176             ;;
 177         "-trace-comment")
 178             TRACE_COMMENT="$2"
 179             shift 2
 180             ;;
 181         "-trace-comment-file")
 182             TRACE_COMMENT_FILE="$2"
 183             shift 2
 184             ;;
 185         "-trace-file")
 186             TRACE_FILENAME="$2"
 187             shift 2
 188             ;;
 189         "-trace-grouped")
 190             TRACE_GROUPED="true"
 191             shift 1
 192             ;;
 193         "-trace-resource")
 194             TRACE_RESOURCE="true"
 195             shift 1
 196             ;;
 197         "-keep-temps")
 198             KEEP="true"
 199             SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes"
 200             shift 1
 201             ;;
 202         "-wrapper")
 203             WRAPPER="$2"
 204             shift 2
 205             ;;
 206         "-help" | "--help" | "-h")
 207             usage
 208             exit 0
 209             ;;
 210         "-version" | "--version" | "-v")
 211             printf '%b\n' "$SIMGRID_VERSION"
 212             exit 0
 213             ;;
 214         "-git-version" | "--git-version")
 215             printf '%b\n' "$SIMGRID_GITHASH"
 216             exit 0
 217             ;;
 218         "--cfg="*|"--log="*)
 219             for OPT in ${1#*=}
 220             do
 221                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 222             done
 223             shift 1
 224             ;;
 225         "-foreground")
 226             # Nothing to do, compatibility.
 227             shift 1
 228             ;;
 229         *)
 230             break
 231             ;;
 232     esac
 233 done
 234
 235 # check if we still have at least one parameter beyond options
 236 if [ $# -eq 0 ]
 237 then
 238     echo "Error: no program to execute!"
 239     usage
 240     exit
 241 fi
 242
 243 EXEC="$1"
 244 shift
 245
 246 # steel --cfg and --logs options
 247 while [ $# -gt 0 ]; do
 248     case "$1" in
 249         "--cfg="*|"--log="*)
 250             for OPT in ${1#*=}
 251             do
 252                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 253             done
 254             shift 1
 255             ;;
 256         *)
 257             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 258             shift
 259             ;;
 260     esac
 261 done
 262
 263 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 264     echo "No hostfile nor platform specified."
 265     usage
 266     exit 1
 267 fi
 268
 269 if [ -z "${HOSTFILE}" ] ; then
 270     HOSTFILETMP=1
 271     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 272     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 273     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) {
 274                 my ($pre,$rad,$post)=($1,$2,$3);
 275                 for my $elm (split(",",$rad)) {
 276                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 277                      for (my $i=$1; $i<=$2;$i++) {
 278                         print "$pre$i$post\n";
 279                      }
 280                   } else {
 281                      print "$pre$elm$post\n";
 282                   }
 283                 }
 284               } elsif (/<cluster/) {
 285             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)");
 286               }' ${PLATFORM} >> ${HOSTFILE}
 287 fi
 288 UNROLLEDHOSTFILETMP=0
 289
 290 #parse if our lines are terminated by :num_process
 291 multiple_processes=$(grep -c ":" $HOSTFILE)
 292 if [ "${multiple_processes}" -gt 0 ] ; then
 293     UNROLLEDHOSTFILETMP=1
 294     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 295     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 296     if [ ${HOSTFILETMP} = 1 ] ; then
 297         rm ${HOSTFILE}
 298         HOSTFILETMP=0
 299     fi
 300     HOSTFILE=$UNROLLEDHOSTFILE
 301 fi
 302
 303 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 304 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
 305 if [ ${hostfile_procs} = 0 ] ; then
 306     die "the hostfile '${HOSTFILE}' is empty"
 307 fi
 308
 309 if [ -z "${NUMPROCS}" ] ; then
 310     # Use the amount of processes in the hostfile as default value for the -np parameter
 311     NUMPROCS=$hostfile_procs
 312 fi
 313
 314 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 315     echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 316 fi
 317
 318 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 319 if [ -z "${PLATFORM}" ]; then
 320     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 321
 322     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 323 <?xml version='1.0'?>
 324 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 325 <platform version="4.1">
 326 <zone id="AS0" routing="Full">
 327 PLATFORMHEAD
 328
 329     i=${NUMPROCS}
 330     while [ $i -gt 0 ]; do
 331         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 332         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 333         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 334         i=$((i - 1))
 335     done
 336
 337     i=${NUMPROCS}
 338     while [ $i -gt 0 ]; do
 339         j=${NUMPROCS}
 340         while [ $j -gt 0 ]; do
 341             if [ $i -eq $j ]; then
 342                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 343             else
 344                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 345             fi
 346             j=$((j - 1))
 347         done
 348         i=$((i - 1))
 349     done
 350
 351     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 352 </zone>
 353 </platform>
 354 PLATFORMFOOT
 355
 356 else
 357     PLATFORMTMP=${PLATFORM}
 358 fi
 359 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 360 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 361 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 362 #APPLICATIONTMP="app.xml"
 363
 364 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 365 <?xml version='1.0'?>
 366 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
 367 <platform version="4.1">
 368 APPLICATIONHEAD
 369
 370 ##---- cache hostnames of hostfile---------------
 371 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 372     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 373 fi
 374
 375 if [ "${EXTOPT}" = "smpi_replay" ]; then
 376     APP_TRACES=$PROC_ARGS
 377     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
 378         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 379         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 380         REPLAY=1
 381     else
 382         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 383         exit 1
 384     fi
 385 fi
 386
 387 ##----------------------------------------------------------
 388 ##  generate application.xml with hostnames from hostfile:
 389 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 390 ##  hostfile has less than i lines.
 391 ##----------------------------------------------------------
 392
 393 HAVE_SEQ="$(which seq 2>/dev/null)"
 394
 395 if [ -n "${HAVE_SEQ}" ]; then
 396     SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
 397 else
 398     cnt=0
 399     while [ $cnt -lt ${NUMPROCS} ] ; do
 400         SEQ="$SEQ $cnt"
 401         cnt=$((cnt + 1));
 402     done
 403 fi
 404
 405 set -- $hostnames
 406
 407 ##---- generate <actor> tags------------------------------
 408 #prepare arguments at once
 409 for ARG in $PROC_ARGS; do
 410   XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
 411 "
 412 done
 413
 414 for i in ${SEQ}
 415 do
 416     j=$(( i % hostfile_procs + 1 ))
 417     host=$(eval "echo \${$j}")
 418
 419     ##---- optional display of ranks to actor mapping
 420     if [ ${MAPOPT} = 1 ]; then
 421       echo "[rank $i] -> $host"
 422     fi
 423
 424     echo "  <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
 425     <argument value=\"smpirun\"/> <!-- instance -->
 426     <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
 427     if [ ${REPLAY} = 1 ]; then
 428         if  [ ${NUMTRACES} -gt 1 ]; then
 429             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 430         else
 431             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 432         fi
 433     else
 434     echo ${XML_ARGS} >> ${APPLICATIONTMP}
 435     fi
 436     echo "  </actor>" >> ${APPLICATIONTMP}
 437 done
 438
 439 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 440 </platform>
 441 APPLICATIONFOOT
 442 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 443 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 444 if [ -n "${TRACE_ACTIVE}" ]; then
 445     #define trace filename
 446     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 447         if [ -z "${TRACE_FILENAME}" ]; then
 448             TRACE_FILENAME="smpi_simgrid.txt"
 449         fi
 450         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 451     else
 452         if [ -z "${TRACE_FILENAME}" ]; then
 453             TRACE_FILENAME="smpi_simgrid.trace"
 454         fi
 455         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 456     fi
 457
 458     if [ -n "${TRACE_COMMENT}" ]; then
 459         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 460     fi
 461
 462     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 463         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 464     fi
 465
 466     if [ -n "${TRACE_GROUPED}" ]; then
 467         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 468     fi
 469
 470     if [ -n "${TRACE_RESOURCE}" ]; then
 471         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 472     fi
 473 fi
 474 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 475
 476 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 477 if [ -n "${KEEP}" ] ; then
 478     echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 479     if [ ${HOSTFILETMP} = 1 ] ; then
 480         echo "Generated hostfile ${HOSTFILE} kept."
 481     fi
 482     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 483         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 484     fi
 485 fi
 486
 487 # Execute the process
 488 #
 489 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 490 #
 491 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 492 #
 493 # * The job is launched in the background in order to be able to handle signals.
 494 #
 495 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 496 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 497 exec 3<&0
 498 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 499 pid=$!
 500 exec 3>&-
 501 wait $pid
 502 status=$?
 503 pid=""
 504
 505 # Keep temporary files on failures to help debugging
 506 #
 507 if [ ${status} -ne 0 ] ; then
 508     if [ -z "${KEEP}" ]; then
 509         echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 510         if [ ${HOSTFILETMP} = 1 ] ; then
 511             echo "Generated hostfile ${HOSTFILE} kept."
 512         fi
 513         if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 514             echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 515         fi
 516     fi
 517     echo "Execution failed with code ${status}."
 518     KEEP=true
 519 fi
 520
 521 smpirun_cleanup
 522
 523 exit $status