X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/6b188a3ca7fe91cf125a08229fb4fe51eee26e6d..adbdcbddadcbd685b72259874d1ebfd9cbbcc497:/src/smpi/smpirun.in diff --git a/src/smpi/smpirun.in b/src/smpi/smpirun.in index ad8295b63e..194934f9e8 100755 --- a/src/smpi/smpirun.in +++ b/src/smpi/smpirun.in @@ -1,36 +1,71 @@ -#!/bin/bash +#!/usr/bin/env sh + +# Copyright (c) 2007-2023. 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. @CMAKE_SMPI_COMMAND@ -DEFAULT_LOOPBACK_BANDWIDTH="498000000" -DEFAULT_LOOPBACK_LATENCY="0.000004" -DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))" -DEFAULT_NETWORK_LATENCY="0.000005" -DEFAULT_NUMPROCS="4" -DEFAULT_POWER="100" +SIMGRID_VERSION="@SIMGRID_VERSION_STRING@" +SIMGRID_GITHASH="@GIT_VERSION@" + +DEFAULT_LOOPBACK_BANDWIDTH="498000000Bps" +DEFAULT_LOOPBACK_LATENCY="0.000004s" +DEFAULT_NETWORK_BANDWIDTH="$((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}" -NUMPROCS="${DEFAULT_NUMPROCS}" -POWER="${DEFAULT_POWER}" +SPEED="${DEFAULT_SPEED}" + +PRIVATIZE="--cfg=smpi/privatization:${SMPI_PRIVATIZATION:-@HAVE_PRIVATIZATION@}" + +NUMPROCS=0 +DEPLOYOPTS="" -SIMOPTS="--cfg=maxmin/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304" +SIMOPTS="--cfg=precision/timing:1e-9 --cfg=network/model:SMPI" + +SMPITMPDIR="$(dirname $(mktemp -u))" #usage to print the way this script should be called -function usage () { - echo "usage:" - echo "$0 [-quiet] [-keep-temps]" - echo " [-np ] -platform -hostfile [-map] program [program-options]" - echo " [-trace] # activate tracing" - echo " [-trace-grouped] # group MPI processes by location" - echo " [-trace-resource] # trace resource utilization" - echo " [-trace-triva] # generate configuration for Triva's GraphView" - echo " [-trace-file ] # name of the tracefile (simgrid_smpi.trace)" - echo "or (deprecated usage):" - echo "$0 [-quiet] [-keep-temps] [-np ] [-bandwidth ] [-latency ] program [program-options]" - echo +usage () { + cat < -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" 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. + -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-file # name of the tracefile (simgrid_smpi.trace) + -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 + -help # Displays this information + -help-coll # Displays all available collective algorithms + +or (deprecated usage): + $0 [-keep-temps] [-np ] [-bandwidth ] [-latency ] program [program-options] + +EOF } #check if we have at least one parameter @@ -40,239 +75,396 @@ then exit fi -while true; do - case "$1" in - "-np" | "-n") - NUMPROCS="$2" - shift 2 - ;; - "-bandwidth") - NETWORK_BANDWIDTH="$2" - shift 2 - ;; - "-latency") - NETWORK_LATENCY="$2" - shift 2 - ;; - "-platform") - PLATFORM="$2" - if [ ! -f ${PLATFORM} ]; then - echo "[$0] ** error: the file '${PLATFORM}' does not exist. Aborting." - exit 1 - fi - shift 2 - ;; - "-hostfile") - HOSTFILE="$2" - if [ ! -f ${HOSTFILE} ]; then - echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting." - exit 1 - fi - shift 2 - ;; - - "-map") - MAPOPT="true" - shift 1 - ;; - - "-trace") - TRACE_ACTIVE="true" - shift 1 - ;; - - "-trace-file") - TRACE_FILENAME="$2" - shift 2 - ;; - - "-trace-grouped") - TRACE_GROUPED="true" - shift 1 - ;; - - "-trace-resource") - TRACE_RESOURCE="true" - shift 1 - ;; - - "-trace-triva") - TRACE_TRIVA="true" - shift 1 - ;; - - "-keep-temps") - KEEP="true" - shift 1 - ;; - - "-help" | "--help" | "-h") - usage - exit - ;; - - "--cfg="*|"--log="*) - for OPT in ${1#*=} - do - SIMOPTS="$SIMOPTS ${1%%=*}=$OPT" - done - shift 1 - ;; - *) - break - ;; - esac -done +WRAPPER="" +HOSTFILE="" +QUIET="" -EXEC="$1" -shift +unset pid -##----------------------------------- +trapped_signals="HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS" -# test if we have something to execute, otherwise show usage and exit -if [ -z ${EXEC} ] -then - echo "Program is missing" - usage +die () { + printf '[%s] ** error: %s. Aborting.\n' "$(basename "$0")" "$*" >&2 exit 1 -fi +} + +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 "$pid" + fi + unset pid -##-------------------------------- DEFAULT or SPECIFIED PLATFORM -------------------------------------- -if [ -z "${PLATFORM}" ]; then - PLATFORMTMP="$(mktemp tmpXXXXXX)" + # Raise the same signal again (remove the traps first): + trap - $trapped_signals + kill -"$sig" $$ - cat > ${PLATFORMTMP} < - - - -PLATFORMHEAD + # This should never happen: + kill -ABRT $$ + kill -TERM $$ +} -for (( i=${NUMPROCS}; $i ; i=$i-1 )) do - echo " " >> ${PLATFORMTMP} - echo " " >> ${PLATFORMTMP} - echo " " >> ${PLATFORMTMP} +for s in $trapped_signals; do + trap "smpirun_trap $s" "$s" done -for (( i=${NUMPROCS}; $i ; i=$i-1 )) do - for (( j=${NUMPROCS}; $j ; j=$j-1 )) do - if [ $i -eq $j ]; then - echo " " >> ${PLATFORMTMP} - else - echo " " >> ${PLATFORMTMP} - fi - done +while true; do + case "$1" in + "-np" | "-n") + NUMPROCS="$2" + shift 2 + ;; + "-bandwidth") + NETWORK_BANDWIDTH="$2" + shift 2 + ;; + "-latency") + NETWORK_LATENCY="$2" + shift 2 + ;; + "-platform") + PLATFORM="$2" + if [ ! -f "${PLATFORM}" ]; then + die "the file '${PLATFORM}' does not exist" + fi + shift 2 + ;; + "-hostfile" | "-machinefile") + HOSTFILE="$2" + if [ ! -f "${HOSTFILE}" ]; then + die "the file '${HOSTFILE}' does not exist" + fi + shift 2 + ;; + "-replay") + APP_TRACES="$2" + if [ ! -f "${APP_TRACES}" ]; then + die "the file '${APP_TRACES}' does not exist" + fi + DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/replay:${APP_TRACES}" + shift 2 + ;; + "-no-privatize") + PRIVATIZE="--cfg=smpi/privatization:no" + shift 1 + ;; + "-map") + DEPLOYOPTS="${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="$2" + shift 2 + ;; + "-trace-grouped") + TRACE_GROUPED="true" + shift 1 + ;; + "-trace-resource") + TRACE_RESOURCE="true" + shift 1 + ;; + "-keep-temps") + KEEP="true" + SIMOPTS="$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 --cfg=smpi/pedantic:true --cfg=smpi/barrier-collectives:true" + shift 1 + ;; + "-help" | "--help" | "-h") + usage + exit 0 + ;; + "-help-coll" | "--help-coll") + ${WRAPPER} "@SMPIMAIN@" --help-coll + exit 0 + ;; + "-version" | "--version" | "-v") + printf '%b\n' "$SIMGRID_VERSION" + exit 0 + ;; + "-git-version" | "--git-version") + printf '%b\n' "$SIMGRID_GITHASH" + exit 0 + ;; + "--cfg="*|"--log="*) + for OPT in ${1#*=} + do + SIMOPTS="$SIMOPTS ${1%%=*}=$OPT" + done + shift 1 + ;; + "-foreground") + # Nothing to do, compatibility. + shift 1 + ;; + *) + break + ;; + esac done -cat >> ${PLATFORMTMP} < - -PLATFORMFOOT +#setup tmp dir +SIMOPTS="$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="$1" + shift + fi else - PLATFORMTMP=${PLATFORM} -fi -##-------------------------------- end DEFAULT or SPECIFIED PLATFORM -------------------------------------- - -##-------------------------------- DEFAULT APPLICATION -------------------------------------- -APPLICATIONTMP="$(mktemp tmpXXXXXX)" -#APPLICATIONTMP="app.xml" - + # check if we still have at least one parameter beyond options + if [ $# -eq 0 ] + then + echo "Error: no program to execute!" + usage + exit + fi -cat > ${APPLICATIONTMP} < - - -APPLICATIONHEAD - -##---- cache hostnames of hostfile--------------- -if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then - hostnames=(`cat ${HOSTFILE} | tr \\\n " "`) - NUMHOSTS=`cat ${HOSTFILE} | wc -l` + EXEC="$1" + shift fi -##---------------------------------------------------------- -## generate application.xml with hostnames from hostfile: -## the name of host_i (1<=i<=p, where -np p) is the line i -## in hostfile (where -hostfile hostfile), or "host$i" if -## hostfile has less than i lines. -##---------------------------------------------------------- +# steal --cfg and --logs options +while [ $# -gt 0 ]; do + case "$1" in + "--cfg="*|"--log="*) + for OPT in ${1#*=} + do + SIMOPTS="$SIMOPTS ${1%%=*}=$OPT" + done + shift 1 + ;; + *) + PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1" + shift + ;; + esac +done -HAVE_SEQ="`which seq 2>/dev/null`" +if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then + echo "No hostfile nor platform specified." + usage + exit 1 +fi -if [ -n "${HAVE_SEQ}" ]; then - SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))` -else - cnt=0 - while (( $cnt < ${NUMPROCS} )) ; do - SEQ="$SEQ $cnt" - cnt=$((cnt + 1)); - done +UNROLLEDHOSTFILETMP=0 + +# parse if our lines are terminated by :num_process +if [ -n "${HOSTFILE}" ] && grep -q ':' "${HOSTFILE}" ; then + UNROLLEDHOSTFILETMP=1 + 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 -##---- generate tags------------------------------ +DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/np:${NUMPROCS}" +DEPLOYOPTS="${DEPLOYOPTS} --cfg=smpi/hostfile:${HOSTFILE}" -for i in ${SEQ} -do - if [ -n "${HOSTFILE}" ]; then - j=$(( $i % ${NUMHOSTS} )) - fi - ##---- optional display of ranks to process mapping - if [ -n "${MAPOPT}" ]; then - echo "[rank $i] -> ${hostnames[$j]}" - fi +##-------------------------------- DEFAULT or SPECIFIED PLATFORM -------------------------------------- +if [ -z "${PLATFORM}" ]; then + PLATFORMTMP="$(mktemp smpitmp-platfXXXXXX)" - if [ -z "${hostnames[$j]}" ]; then - host="host"$(($j+1)) - else - host="${hostnames[$j]}" - fi - echo " " >> ${APPLICATIONTMP} - echo " " >> ${APPLICATIONTMP} - for ARG in $*; do - echo " " >> ${APPLICATIONTMP} - done - echo " " >> ${APPLICATIONTMP} -done + cat > "${PLATFORMTMP}" < + + + +PLATFORMHEAD -cat >> ${APPLICATIONTMP} <" + echo " " + echo " " + } >> "${PLATFORMTMP}" + i=$((i - 1)) + done + + i=${NUMPROCS} + while [ "$i" -gt 0 ]; do + j=${NUMPROCS} + while [ "$j" -gt 0 ]; do + if [ "$i" -eq "$j" ]; then + echo " " >> "${PLATFORMTMP}" + else + echo " " >> "${PLATFORMTMP}" + fi + j=$((j - 1)) + done + i=$((i - 1)) + done + + cat >> "${PLATFORMTMP}" < -APPLICATIONFOOT -##-------------------------------- end DEFAULT APPLICATION -------------------------------------- +PLATFORMFOOT +else + PLATFORMTMP=${PLATFORM} +fi +##-------------------------------- end DEFAULT or SPECIFIED PLATFORM -------------------------------------- ##---------------------- SMPI TRACING OPTIONS --------------------------------- if [ -n "${TRACE_ACTIVE}" ]; then - #define trace filename - if [ -z "${TRACE_FILENAME}" ]; then - TRACE_FILENAME="smpi_simgrid.trace" - fi - TRACEOPTIONS="--cfg=tracing:1 --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:1" + #define trace filename + if [ -n "${TRACE_TI_ACTIVE}" ]; then + if [ -z "${TRACE_FILENAME}" ]; then + TRACE_FILENAME="smpi_simgrid.txt" + fi + TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes" + else + if [ -z "${TRACE_FILENAME}" ]; then + TRACE_FILENAME="smpi_simgrid.trace" + fi + TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes" + fi - if [ -n "${TRACE_GROUPED}" ]; then - TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:1" - fi + if [ -n "${TRACE_COMMENT}" ]; then + TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}" + fi - if [ -n "${TRACE_RESOURCE}" ]; then - TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:1 --cfg=tracing/uncategorized:1" - fi + if [ -n "${TRACE_COMMENT_FILE}" ]; then + TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment-file:${TRACE_COMMENT_FILE}" + fi - if [ -n "${TRACE_TRIVA}" ]; then - TRACEOPTIONS="${TRACEOPTIONS} --cfg=triva/categorized:smpi_cat.plist --cfg=triva/uncategorized:smpi_uncat.plist" - fi + if [ -n "${TRACE_GROUPED}" ]; then + TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes" + fi + + if [ -n "${TRACE_RESOURCE}" ]; then + TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes" + fi fi ##---------------------- end SMPI TRACING OPTIONS --------------------------------- +# Do not remove, this variable may be used by user code (e.g. StarPU) export SMPI_GLOBAL_SIZE=${NUMPROCS} -if [ -n "${KEEP}" ] ; then - echo ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP} +if [ -n "${KEEP}" ] && [ -z "${QUIET}" ] ; then + echo "${EXEC}" ${PRIVATIZE} "${TRACEOPTIONS}" "${SIMOPTS}" "${PLATFORMTMP}" + if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then + echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} kept." + fi fi -${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP} -if [ -z "${KEEP}" ] ; then - if [ -z "${PLATFORM}" ]; then - rm ${PLATFORMTMP} - fi - rm ${APPLICATIONTMP} +# Execute the process +# +# The shell still need to be alive for the duration in order to do some cleanup after the process. +# +# We are going through great lengths in order to both keep stdin and be able to handle signals: +# +# * The job is launched in the background in order to be able to handle signals. +# +# * The FD 3 is used to temporarily store FD 1. This is because the shell connects FD 1 to /dev/null when the command +# is launched in the background: this can be overridden in bash but not in standard bourne shell. +exec 3<&0 +${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 + echo "Execution failed with code ${status}." fi + +smpirun_cleanup + +exit $status