src/smpi/smpirun.in

   1 #!/usr/bin/env sh
   2
   3 # Copyright (c) 2007-2018. 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:${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        $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> -replay <tracefile> [program] [program-options]
  35 Options:
  36   -keep-temps                # don't remove the generated files after execution
  37   -wrapper <command>         # use command to run the program (e.g. "valgrind" or "gdb --args")
  38   -map                       # display the machine on which each process rank is mapped
  39   -np <numprocs>             # use that amount of processes from the hostfile.
  40                              # By default, all processes of the hostfile are used.
  41   -no-privatize              # Disable the globals privatization, that is activated by default
  42   -trace-ti                  # activate time independant tracing (for replay, default in smpi_simgrid.txt)
  43   -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
  44   -trace-comment <comment>   # put a comment on the top of the trace file
  45   -trace-comment-file <file> # put file contents on the top of the trace file as comment
  46   -trace-grouped             # group MPI processes by location
  47   -trace-resource            # trace resource utilization
  48   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
  49   -replay <tracefile>        # replays a trace instead of actually executing an application
  50
  51   -version                   # Displays the SimGrid version (human readable)
  52   -git-version               # Displays the git hash of SimGrid
  53
  54 or (deprecated usage):
  55   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
  56
  57 EOF
  58 }
  59
  60 #check if we have at least one parameter
  61 if [ $# -eq 0 ]
  62 then
  63     usage
  64     exit
  65 fi
  66
  67 WRAPPER=""
  68 HOSTFILE=""
  69 HOSTFILETMP=0
  70 MAPOPT=0
  71 REPLAY=0
  72
  73 unset pid
  74
  75 trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS"
  76
  77 die () {
  78     printf '[%s] ** error: %s. Aborting.\n' "$(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         "-replay")
 157             APP_TRACES="$2"
 158             shift 2
 159             ;;
 160         "-no-privatize")
 161             PRIVATIZE="--cfg=smpi/privatization:no"
 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 if [ -n "${APP_TRACES}" ] ; then
 236     if [ $# -eq 0 ] ; then
 237         EXEC="@SMPIREPLAYMAIN@"
 238     else
 239         EXEC="$1"
 240         shift
 241     fi
 242 else
 243     # check if we still have at least one parameter beyond options
 244     if [ $# -eq 0 ]
 245     then
 246         echo "Error: no program to execute!"
 247         usage
 248         exit
 249     fi
 250
 251     EXEC="$1"
 252     shift
 253 fi
 254
 255 # steel --cfg and --logs options
 256 while [ $# -gt 0 ]; do
 257     case "$1" in
 258         "--cfg="*|"--log="*)
 259             for OPT in ${1#*=}
 260             do
 261                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
 262             done
 263             shift 1
 264             ;;
 265         *)
 266             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
 267             shift
 268             ;;
 269     esac
 270 done
 271
 272 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
 273     echo "No hostfile nor platform specified."
 274     usage
 275     exit 1
 276 fi
 277
 278 if [ -z "${HOSTFILE}" ] ; then
 279     HOSTFILETMP=1
 280     HOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 281     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
 282     # put all <cluster tag on its own line.
 283     cat ${PLATFORM} | tr '\n' ' ' | sed 's/<cluster/\n<cluster/' | \
 284       perl -ne 'if (m/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/s) {
 285                 my ($pre,$rad,$post)=($1,$2,$3);
 286                 for my $elm (split(",",$rad)) {
 287                   if ($elm=~/^([^-]*?)-([^-]*)$/) {
 288                      for (my $i=$1; $i<=$2;$i++) {
 289                         print "$pre$i$post\n";
 290                      }
 291                   } else {
 292                      print "$pre$elm$post\n";
 293                   }
 294                 }
 295               } elsif (/<cluster/) {
 296             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.");
 297               }' ${PLATFORM} >> ${HOSTFILE}
 298 fi
 299 UNROLLEDHOSTFILETMP=0
 300
 301 #parse if our lines are terminated by :num_process
 302 multiple_processes=$(grep -c ":" $HOSTFILE)
 303 if [ "${multiple_processes}" -gt 0 ] ; then
 304     UNROLLEDHOSTFILETMP=1
 305     UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)"
 306     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
 307     if [ ${HOSTFILETMP} = 1 ] ; then
 308         rm ${HOSTFILE}
 309         HOSTFILETMP=0
 310     fi
 311     HOSTFILE=$UNROLLEDHOSTFILE
 312 fi
 313
 314 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
 315 hostfile_procs=$(grep -c "[a-zA-Z0-9]" $HOSTFILE)
 316 if [ ${hostfile_procs} = 0 ] ; then
 317     die "the hostfile '${HOSTFILE}' is empty"
 318 fi
 319
 320 if [ -z "${NUMPROCS}" ] ; then
 321     # Use the amount of processes in the hostfile as default value for the -np parameter
 322     NUMPROCS=$hostfile_procs
 323 fi
 324
 325 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
 326     echo "You requested to use ${NUMPROCS} ranks, but there is only ${hostfile_procs} processes in your hostfile..." >&2
 327 fi
 328
 329 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 330 if [ -z "${PLATFORM}" ]; then
 331     PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)"
 332
 333     cat > ${PLATFORMTMP} <<PLATFORMHEAD
 334 <?xml version='1.0'?>
 335 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
 336 <platform version="4.1">
 337 <zone id="AS0" routing="Full">
 338 PLATFORMHEAD
 339
 340     i=${NUMPROCS}
 341     while [ $i -gt 0 ]; do
 342         echo "  <host id=\"host$i\" speed=\"${SPEED}\"/>" >> ${PLATFORMTMP}
 343         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
 344         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
 345         i=$((i - 1))
 346     done
 347
 348     i=${NUMPROCS}
 349     while [ $i -gt 0 ]; do
 350         j=${NUMPROCS}
 351         while [ $j -gt 0 ]; do
 352             if [ $i -eq $j ]; then
 353                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
 354             else
 355                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
 356             fi
 357             j=$((j - 1))
 358         done
 359         i=$((i - 1))
 360     done
 361
 362     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 363 </zone>
 364 </platform>
 365 PLATFORMFOOT
 366
 367 else
 368     PLATFORMTMP=${PLATFORM}
 369 fi
 370 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 371 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
 372 APPLICATIONTMP="$(mktemp smpitmp-appXXXXXX)"
 373 #APPLICATIONTMP="app.xml"
 374
 375 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
 376 <?xml version='1.0'?>
 377 <!DOCTYPE platform SYSTEM "https://simgrid.org/simgrid.dtd">
 378 <platform version="4.1">
 379 APPLICATIONHEAD
 380
 381 ##---- cache hostnames of hostfile---------------
 382 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
 383     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
 384 fi
 385
 386 if [ -n "${APP_TRACES}" ]; then
 387     if [ -f "${APP_TRACES}" ]; then
 388         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
 389         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
 390         REPLAY=1
 391     else
 392         printf "File not found: %s\n" "${APP_TRACES:-\${APP_TRACES\}}" >&2
 393         exit 1
 394     fi
 395 fi
 396
 397 ##----------------------------------------------------------
 398 ##  generate application.xml with hostnames from hostfile:
 399 ##  the name of host_i (1<=i<=p, where -np p) is the line i in hostfile (where -hostfile hostfile), or "host$i" if
 400 ##  hostfile has less than i lines.
 401 ##----------------------------------------------------------
 402
 403 HAVE_SEQ="$(which seq 2>/dev/null)"
 404
 405 if [ -n "${HAVE_SEQ}" ]; then
 406     SEQ=$(${HAVE_SEQ} 0 $(( NUMPROCS - 1)))
 407 else
 408     cnt=0
 409     while [ $cnt -lt ${NUMPROCS} ] ; do
 410         SEQ="$SEQ $cnt"
 411         cnt=$((cnt + 1));
 412     done
 413 fi
 414
 415 set -- $hostnames
 416
 417 ##---- generate <actor> tags------------------------------
 418 #prepare arguments at once
 419 for ARG in $PROC_ARGS; do
 420   XML_ARGS="${XML_ARGS}""<argument value=\"${ARG}\"/>
 421 "
 422 done
 423
 424 for i in ${SEQ}
 425 do
 426     j=$(( i % hostfile_procs + 1 ))
 427     host=$(eval "echo \${$j}")
 428
 429     ##---- optional display of ranks to actor mapping
 430     if [ ${MAPOPT} = 1 ]; then
 431       echo "[rank $i] -> $host"
 432     fi
 433
 434     echo "  <actor host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->
 435     <prop id=\"instance_id\" value=\"smpirun\"/>
 436     <prop id=\"rank\" value=\"$i\"/>" >> ${APPLICATIONTMP}
 437     if [ ${REPLAY} = 1 ]; then
 438         echo "    <prop id=\"smpi_replay\" value=\"true\"/>" >> ${APPLICATIONTMP}
 439         if  [ ${NUMTRACES} -gt 1 ]; then
 440             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
 441         else
 442             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
 443         fi
 444     else
 445     echo ${XML_ARGS} >> ${APPLICATIONTMP}
 446     fi
 447     echo "  </actor>" >> ${APPLICATIONTMP}
 448 done
 449
 450 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
 451 </platform>
 452 APPLICATIONFOOT
 453 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
 454 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
 455 if [ -n "${TRACE_ACTIVE}" ]; then
 456     #define trace filename
 457     if [ -n "${TRACE_TI_ACTIVE}" ]; then
 458         if [ -z "${TRACE_FILENAME}" ]; then
 459             TRACE_FILENAME="smpi_simgrid.txt"
 460         fi
 461         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
 462     else
 463         if [ -z "${TRACE_FILENAME}" ]; then
 464             TRACE_FILENAME="smpi_simgrid.trace"
 465         fi
 466         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
 467     fi
 468
 469     if [ -n "${TRACE_COMMENT}" ]; then
 470         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
 471     fi
 472
 473     if [ -n "${TRACE_COMMENT_FILE}" ]; then
 474         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}"
 475     fi
 476
 477     if [ -n "${TRACE_GROUPED}" ]; then
 478         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
 479     fi
 480
 481     if [ -n "${TRACE_RESOURCE}" ]; then
 482         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
 483     fi
 484 fi
 485 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
 486
 487 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 488 if [ -n "${KEEP}" ] ; then
 489     echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 490     if [ ${HOSTFILETMP} = 1 ] ; then
 491         echo "Generated hostfile ${HOSTFILE} kept."
 492     fi
 493     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 494         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 495     fi
 496 fi
 497
 498 # Execute the process
 499 #
 500 # The shell still need to be alive for the duration in order to do some cleanup after the process.
 501 #
 502 # We are going through great lengths in order to both keep stdin and be able to handle signals:
 503 #
 504 # * The job is launched in the background in order to be able to handle signals.
 505 #
 506 # * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command
 507 #   is launched in the background: this can be overriden in bash but not in standard bourne shell.
 508 exec 3<&0
 509 ${WRAPPER} "@SMPIMAIN@" ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} <&3 3>&- &
 510 pid=$!
 511 exec 3>&-
 512 wait $pid
 513 status=$?
 514 # With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit
 515 # status of 128. Try again.
 516 while test $status -eq 128 && kill -0 $pid 2>/dev/null; do
 517     wait $pid
 518     status=$?
 519 done
 520 pid=""
 521
 522 # Keep temporary files on failures to help debugging
 523 #
 524 if [ ${status} -ne 0 ] ; then
 525     if [ -z "${KEEP}" ]; then
 526         echo ${EXEC} ${PRIVATIZE} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
 527         if [ ${HOSTFILETMP} = 1 ] ; then
 528             echo "Generated hostfile ${HOSTFILE} kept."
 529         fi
 530         if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
 531             echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept."
 532         fi
 533     fi
 534     echo "Execution failed with code ${status}."
 535     KEEP=true
 536 fi
 537
 538 smpirun_cleanup
 539
 540 exit $status