X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/ffe8ce65fd9a8e18a0469f26f067c3ea6d5d60d4..73e97710413bba3ee2ae8baab0537fbd78811016:/src/smpi/smpirun.in diff --git a/src/smpi/smpirun.in b/src/smpi/smpirun.in index 67e4624d23..8b99fb69c1 100755 --- a/src/smpi/smpirun.in +++ b/src/smpi/smpirun.in @@ -1,7 +1,6 @@ -#! /bin/sh +#!/usr/bin/env sh -# Copyright (c) 2007-2014. The SimGrid Team. -# All rights reserved. +# Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. # This program is free software; you can redistribute it and/or modify it # under the terms of the license (GNU LGPL) which comes with this package. @@ -9,41 +8,57 @@ @CMAKE_SMPI_COMMAND@ SIMGRID_VERSION="@SIMGRID_VERSION_STRING@" +SIMGRID_GITHASH="@SIMGRID_GITHASH@" -DEFAULT_LOOPBACK_BANDWIDTH="498000000" -DEFAULT_LOOPBACK_LATENCY="0.000004" -DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))" -DEFAULT_NETWORK_LATENCY="0.000005" -DEFAULT_NUMPROCS="4" -DEFAULT_POWER="100" +DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps" +DEFAULT_LOOPBACK_LATENCY="0.000004s" +DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))Bps" +DEFAULT_NETWORK_LATENCY="0.000005s" +DEFAULT_SPEED="100flops" LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}" LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}" NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}" NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}" -POWER="${DEFAULT_POWER}" +SPEED="${DEFAULT_SPEED}" -SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304" +PRIVATIZE="--cfg=smpi/privatization:${SMPI_PRIVATIZATION:-@HAVE_PRIVATIZATION@}" + +NUMPROCS=0 +DEPLOYOPTS="" + +SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI" + +SMPITMPDIR="$(dirname $(mktemp -u))" #usage to print the way this script should be called usage () { cat < -hostfile program [program-options] +Usage: $0 [OPTIONS] -platform -hostfile program [program-options] + $0 [OPTIONS] -platform -hostfile -replay [program] [program-options] Options: + -analyze # show information about allocations and timings at the end of execution -keep-temps # don't remove the generated files after execution - -wrapper # use command to run the program (e.g. "valgrind") + -wrapper # use command to run the program (e.g. "valgrind" or "gdb --args") + -gdb # run within GDB (-wrapper "gdb --args" -keep-temps) + -lldb # run within LLDB (-wrapper "lldb --" -keep-temps) + -vgdb # run within Valgrind+GDB (-wrapper "valgrind --vgdb=yes --vgdb-error=0" -keep-temps) -map # display the machine on which each process rank is mapped -np # use that amount of processes from the hostfile. # By default, all processes of the hostfile are used. - -trace-ti # activate time independant tracing (for replay, default in smpi_simgrid.txt) + -no-privatize # Disable the globals privatization, that is activated by default + -tmpdir # Directory used to store temporary files. Defaults to system's. + -trace-ti # activate time independent tracing (for replay, default in smpi_simgrid.txt) -trace # activate tracing (Paje, default in smpi_simgrid.trace) -trace-comment # put a comment on the top of the trace file -trace-comment-file # put file contents on the top of the trace file as comment -trace-grouped # group MPI processes by location -trace-resource # trace resource utilization - -trace-viva # generate configuration for Viva's GraphView -trace-file # name of the tracefile (simgrid_smpi.trace) - -ext # additional parameter (reserved) + -replay # replays a trace instead of actually executing an application + + -version # Displays the SimGrid version (human readable) + -git-version # Displays the git hash of SimGrid or (deprecated usage): $0 [-keep-temps] [-np ] [-bandwidth ] [-latency ] program [program-options] @@ -58,9 +73,56 @@ then exit fi -EXTOPT="" WRAPPER="" HOSTFILE="" +QUIET="" + +unset pid + +trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS" + +die () { + printf '[%s] ** error: %s. Aborting.\n' "$(basename "$0")" "$*" >&2 + exit 1 +} + +smpirun_cleanup() +{ + if [ -z "${KEEP}" ] ; then + if [ -z "${PLATFORM}" ] && [ -n "$PLATFORMTMP" ]; then + rm -f "${PLATFORMTMP}" + PLATFORMTMP="" + fi + if [ "${UNROLLEDHOSTFILETMP}" = 1 ] && [ -n "$UNROLLEDHOSTFILE" ] ; then + rm -f "${UNROLLEDHOSTFILE}" + UNROLLEDHOSTFILE="" + fi + fi +} + +smpirun_trap() { + local sig + sig="$1" + + # Cleanup and kill the child process: + smpirun_cleanup + if [ -n "$pid" ]; then + kill -TERM "$pid" + fi + unset pid + + # Raise the same signal again (remove the traps first): + trap - $trapped_signals + kill -"$sig" $$ + + # This should never happen: + kill -ABRT $$ + kill -TERM $$ +} + +for s in $trapped_signals; do + trap "smpirun_trap $s" "$s" +done while true; do case "$1" in @@ -77,102 +139,115 @@ while true; do shift 2 ;; "-platform") - PLATFORM="$2" + PLATFORM="$2" if [ ! -f "${PLATFORM}" ]; then - echo "[`basename $0`] ** error: the file '${PLATFORM}' does not exist. Aborting." - exit 1 + die "the file '${PLATFORM}' does not exist" fi - shift 2 + shift 2 ;; - "-hostfile") - HOSTFILE="$2" + "-hostfile" | "-machinefile") + HOSTFILE="$2" if [ ! -f "${HOSTFILE}" ]; then - echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting." - exit 1 + die "the file '${HOSTFILE}' does not exist" fi - shift 2 + shift 2 ;; - - "-machinefile") - HOSTFILE="$2" - if [ ! -f "${HOSTFILE}" ]; then - echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting." - exit 1 + "-replay") + APP_TRACES="$2" + if [ ! -f "${APP_TRACES}" ]; then + die "the file '${APP_TRACES}' does not exist" fi - shift 2 - ;; - - "-ext") - EXTOPT="$2" + DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/replay:${APP_TRACES}" shift 2 ;; - + "-no-privatize") + PRIVATIZE="--cfg=smpi/privatization:no" + shift 1 + ;; "-map") - MAPOPT="true" + DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/map:1" + shift 1 + ;; + "-tmpdir") + SMPITMPDIR="$2" shift 1 ;; - "-trace") TRACE_ACTIVE="true" shift 1 ;; - "-trace-ti") TRACE_ACTIVE="true" TRACE_TI_ACTIVE="true" shift 1 ;; - "-trace-comment") TRACE_COMMENT="$2" shift 2 ;; - "-trace-comment-file") TRACE_COMMENT_FILE="$2" shift 2 ;; - "-trace-file") TRACE_FILENAME="$2" shift 2 ;; - "-trace-grouped") TRACE_GROUPED="true" shift 1 ;; - "-trace-resource") TRACE_RESOURCE="true" shift 1 ;; - - "-trace-viva") - TRACE_VIVA="true" - shift 1 - ;; - "-keep-temps") - KEEP="true" + KEEP="true" + SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" + shift 1 + ;; + "-quiet") + QUIET="true" shift 1 ;; - "-wrapper") WRAPPER="$2" shift 2 ;; - + "-gdb") + WRAPPER="gdb --args" + KEEP="true" + SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" + shift 1 + ;; + "-vgdb") + WRAPPER="valgrind --vgdb=yes --vgdb-error=0" + KEEP="true" + SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" + shift 1 + ;; + "-lldb") + WRAPPER="lldb --" + KEEP="true" + SIMOPTS="$SIMOPTS --cfg=smpi/keep-temps:yes" + shift 1 + ;; + "-analyze") + SIMOPTS="$SIMOPTS --cfg=smpi/display-timing:yes --cfg=smpi/display-allocs:yes --cfg=smpi/list-leaks:50" + shift 1 + ;; "-help" | "--help" | "-h") usage exit 0 ;; - "-version" | "--version" | "-v") printf '%b\n' "$SIMGRID_VERSION" exit 0 ;; - + "-git-version" | "--git-version") + printf '%b\n' "$SIMGRID_GITHASH" + exit 0 + ;; "--cfg="*|"--log="*) for OPT in ${1#*=} do @@ -180,20 +255,41 @@ while true; do done shift 1 ;; + "-foreground") + # Nothing to do, compatibility. + shift 1 + ;; *) break ;; esac done -if [ -n "$WRAPPER" ]; then - EXEC="$WRAPPER $1" +#setup tmp dir +SIMOPTS="$SIMOPTS --cfg=smpi/tmpdir:$SMPITMPDIR" +export LD_LIBRARY_PATH="$SMPITMPDIR:$LD_LIBRARY_PATH" + +if [ -n "${APP_TRACES}" ] ; then + if [ $# -eq 0 ] ; then + EXEC="@SMPIREPLAYMAIN@" + else + EXEC="$1" + shift + fi else + # check if we still have at least one parameter beyond options + if [ $# -eq 0 ] + then + echo "Error: no program to execute!" + usage + exit + fi + EXEC="$1" + shift fi -shift -# steel --cfg and --logs options +# steal --cfg and --logs options while [ $# -gt 0 ]; do case "$1" in "--cfg="*|"--log="*) @@ -205,95 +301,78 @@ while [ $# -gt 0 ]; do ;; *) PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1" - shift + shift ;; esac done - -##----------------------------------- - - if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then echo "No hostfile nor platform specified." usage exit 1 fi -HOSTFILETMP=0 -if [ -z "${HOSTFILE}" ] ; then - HOSTFILETMP=1 - HOSTFILE="$(mktemp tmphostXXXXXX)" - perl -ne 'print "$1\n" if /.*.*/' ${PLATFORM} > ${HOSTFILE} -fi UNROLLEDHOSTFILETMP=0 -#parse if our lines are terminated by :num_process -multiple_processes=`grep -c ":" $HOSTFILE` -if [ "${multiple_processes}" -gt 0 ] ; then +# parse if our lines are terminated by :num_process +if [ -n "${HOSTFILE}" ] && grep -q ':' "${HOSTFILE}" ; then UNROLLEDHOSTFILETMP=1 - UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)" - perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/' ${HOSTFILE} > ${UNROLLEDHOSTFILE} - if [ ${HOSTFILETMP} = 1 ] ; then - rm ${HOSTFILE} - HOSTFILETMP=0 - fi + UNROLLEDHOSTFILE="$(mktemp smpitmp-hostfXXXXXX)" + @PYTHON_EXECUTABLE@ -c ' +import sys +import re + +for line in sys.stdin: + m = re.match("(.*):(.*)", line) + if m: + for i in range(0, int(m.group(2))): + print(m.group(1)) + else: + print(line.strip()) +' < "${HOSTFILE}" > "${UNROLLEDHOSTFILE}" HOSTFILE=$UNROLLEDHOSTFILE fi - -# Don't use wc -l to compute it to avoid issues with trailing \n at EOF -hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE` -if [ ${hostfile_procs} = 0 ] ; then - echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2 - exit 1 -fi - -if [ -z "${NUMPROCS}" ] ; then - # Use the amount of processes in the hostfile as default value for the -np parameter - NUMPROCS=$hostfile_procs -fi - - -if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then - echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2 -fi +DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/np:${NUMPROCS}" +DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/hostfile:${HOSTFILE}" ##-------------------------------- DEFAULT or SPECIFIED PLATFORM -------------------------------------- if [ -z "${PLATFORM}" ]; then - PLATFORMTMP="$(mktemp tmpXXXXXX)" + PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)" - cat > ${PLATFORMTMP} < "${PLATFORMTMP}" < - - - + + + PLATFORMHEAD i=${NUMPROCS} - while [ $i -gt 0 ]; do - echo " " >> ${PLATFORMTMP} - echo " " >> ${PLATFORMTMP} - echo " " >> ${PLATFORMTMP} + while [ "$i" -gt 0 ]; do + { + echo " " + echo " " + echo " " + } >> "${PLATFORMTMP}" i=$((i - 1)) done i=${NUMPROCS} - while [ $i -gt 0 ]; do + while [ "$i" -gt 0 ]; do j=${NUMPROCS} - while [ $j -gt 0 ]; do - if [ $i -eq $j ]; then - echo " " >> ${PLATFORMTMP} + while [ "$j" -gt 0 ]; do + if [ "$i" -eq "$j" ]; then + echo " " >> "${PLATFORMTMP}" else - echo " " >> ${PLATFORMTMP} + echo " " >> "${PLATFORMTMP}" fi j=$((j - 1)) done i=$((i - 1)) done - cat >> ${PLATFORMTMP} < + cat >> "${PLATFORMTMP}" < PLATFORMFOOT @@ -301,93 +380,6 @@ else PLATFORMTMP=${PLATFORM} fi ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM -------------------------------------- - -##-------------------------------- DEFAULT APPLICATION -------------------------------------- -APPLICATIONTMP="$(mktemp tmpXXXXXX)" -#APPLICATIONTMP="app.xml" - - -cat > ${APPLICATIONTMP} < - - -APPLICATIONHEAD - -##---- cache hostnames of hostfile--------------- -if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then - hostnames=$(cat ${HOSTFILE} | tr '\n\r' ' ') - NUMHOSTS=$(cat ${HOSTFILE} | wc -l) -fi - -if [ "${EXTOPT}" = "smpi_replay" ]; then - APP_TRACES=$PROC_ARGS - if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then - hosttraces=$(cat ${APP_TRACES} | tr '\n\r' ' ' ) - NUMTRACES=$(cat ${APP_TRACES} | wc -l) - else - printf "File not found: %s\n", "${APP_TRACES:-\${APP_TRACES\}}" >&2 - exit 1 - fi -fi - -##---------------------------------------------------------- -## generate application.xml with hostnames from hostfile: -## the name of host_i (1<=i<=p, where -np p) is the line i -## in hostfile (where -hostfile hostfile), or "host$i" if -## hostfile has less than i lines. -##---------------------------------------------------------- - -HAVE_SEQ="`which seq 2>/dev/null`" - -if [ -n "${HAVE_SEQ}" ]; then - SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))` -else - cnt=0 - while [ $cnt -lt ${NUMPROCS} ] ; do - SEQ="$SEQ $cnt" - cnt=$((cnt + 1)); - done -fi - -##---- generate tags------------------------------ - -for i in ${SEQ} -do - if [ -n "${HOSTFILE}" ]; then - j=$(( $i % ${NUMHOSTS} + 1 )) - fi - ##---- optional display of ranks to process mapping - if [ -n "${MAPOPT}" ]; then - echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)" - fi - - if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then - host="host"$($j) - else - host="$(echo $hostnames|cut -d' ' -f$j)" - fi - echo " " >> ${APPLICATIONTMP} - echo " " >> ${APPLICATIONTMP} - echo " " >> ${APPLICATIONTMP} - if [ "${EXTOPT}" = "smpi_replay" ]; then - if [ ${NUMTRACES} -gt 1 ]; then - echo " " >> ${APPLICATIONTMP} - else - echo " " >> ${APPLICATIONTMP} - fi - else - for ARG in $PROC_ARGS; do - echo " " >> ${APPLICATIONTMP} - done - fi - echo " " >> ${APPLICATIONTMP} -done - -cat >> ${APPLICATIONTMP} < -APPLICATIONFOOT -##-------------------------------- end DEFAULT APPLICATION -------------------------------------- - ##---------------------- SMPI TRACING OPTIONS --------------------------------- if [ -n "${TRACE_ACTIVE}" ]; then #define trace filename @@ -408,7 +400,7 @@ if [ -n "${TRACE_ACTIVE}" ]; then fi if [ -n "${TRACE_COMMENT_FILE}" ]; then - TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}" + TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}" fi if [ -n "${TRACE_GROUPED}" ]; then @@ -418,37 +410,55 @@ if [ -n "${TRACE_ACTIVE}" ]; then if [ -n "${TRACE_RESOURCE}" ]; then TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes" fi - - if [ -n "${TRACE_VIVA}" ]; then - TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist" - fi fi ##---------------------- end SMPI TRACING OPTIONS --------------------------------- +# Do not remove, this variable may be used by user code (e.g. StarPU) export SMPI_GLOBAL_SIZE=${NUMPROCS} -if [ -n "${KEEP}" ] ; then - echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} - if [ ${HOSTFILETMP} = 1 ] ; then - echo "Generated hostfile ${HOSTFILE} keeped." - fi +if [ -n "${KEEP}" ] && [ -z "${QUIET}" ] ; then + echo "${EXEC}" ${PRIVATIZE} "${TRACEOPTIONS}" "${SIMOPTS}" "${PLATFORMTMP}" if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then - echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped." + echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept." fi fi -${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} -status=$? -if [ -z "${KEEP}" ] ; then - if [ -z "${PLATFORM}" ]; then - rm ${PLATFORMTMP} - fi - if [ ${HOSTFILETMP} = 1 ] ; then - rm ${HOSTFILE} - fi - if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then - rm ${UNROLLEDHOSTFILE} +# Execute the process +# +# The shell still need to be alive for the duration in order to do some cleanup after the process. +# +# We are going through great lengths in order to both keep stdin and be able to handle signals: +# +# * The job is launched in the background in order to be able to handle signals. +# +# * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command +# is launched in the background: this can be overridden in bash but not in standard bourne shell. +exec 3<&0 +${WRAPPER} "@SMPIMAIN@" "${EXEC}" ${PRIVATIZE} ${DEPLOYOPTS} ${TRACEOPTIONS} ${SIMOPTS} "${PLATFORMTMP}" ${PROC_ARGS} <&3 3>&- & +pid=$! +exec 3>&- +wait $pid +status=$? +# With dash on Windows WSL/Ubuntu, "wait" sometimes returns early with an exit +# status of 128. Try again. +while test $status -eq 128 && kill -0 $pid 2>/dev/null; do + wait $pid + status=$? +done +pid="" + +# Keep temporary files on failures to help debugging +# +if [ ${status} -ne 0 ] ; then + if [ -z "${KEEP}" ] && [ -z "${QUIET}" ]; then + echo "${EXEC}" ${PRIVATIZE} "${TRACEOPTIONS}" "${SIMOPTS}" "${PLATFORMTMP}" + if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then + echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept." + fi + KEEP=true fi - rm ${APPLICATIONTMP} + echo "Execution failed with code ${status}." fi +smpirun_cleanup + exit $status