Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
dividing by 0 is not always the best possible choice, even if the user shoulnt give...
[simgrid.git] / src / smpi / smpirun.in
index ad8295b..9dfb521 100755 (executable)
@@ -1,7 +1,15 @@
-#!/bin/bash
+#! /bin/sh
+
+# Copyright (c) 2007-2014. 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@
 
+SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
+
 DEFAULT_LOOPBACK_BANDWIDTH="498000000"
 DEFAULT_LOOPBACK_LATENCY="0.000004"
 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
@@ -13,24 +21,34 @@ LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
-NUMPROCS="${DEFAULT_NUMPROCS}"
 POWER="${DEFAULT_POWER}"
 
-SIMOPTS="--cfg=maxmin/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
+SIMOPTS="--cfg=maxmin/precision:1e-3 --cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
 
 #usage to print the way this script should be called
-function usage () {
-    echo "usage:"
-    echo "$0 [-quiet] [-keep-temps]"
-    echo "      [-np <numprocs>] -platform <xmldesc> -hostfile <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 <tracefile>] # name of the tracefile (simgrid_smpi.trace)"
-    echo "or (deprecated usage):"
-    echo "$0 [-quiet] [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]"
-    echo
+usage () {
+    cat <<EOF
+Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
+Options:
+  -keep-temps                # don't remove the generated files after execution
+  -wrapper <command>         # use command to run the program (e.g. "valgrind")
+  -map                       # display the machine on which each process rank is mapped
+  -np <numprocs>             # 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)
+  -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
+  -trace-comment <comment>   # put a comment on the top of the trace file
+  -trace-comment-file <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 <tracefile>    # name of the tracefile (simgrid_smpi.trace)
+  -ext <value>               # additional parameter (reserved)
+
+or (deprecated usage):
+  $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
+
+EOF
 }
 
 #check if we have at least one parameter
@@ -40,139 +58,243 @@ then
     exit
 fi
 
+EXTOPT=""
+WRAPPER=""
+HOSTFILE=""
+
 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
+    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
+            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
+            fi
+           shift 2
+            ;;
+
+        "-machinefile")
+           HOSTFILE="$2"
+            if [ ! -f "${HOSTFILE}" ]; then
+               echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
+               exit 1
+            fi
+           shift 2
+            ;;
+
+        "-ext")
+            EXTOPT="$2"
+            shift 2
+            ;;
+
+        "-map")
+           MAPOPT="true"
+            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"
+            shift 1
+            ;;
+
+        "-wrapper")
+            WRAPPER="$2"
+            shift 2
+            ;;
+
+        "-help" | "--help" | "-h")
+            usage
+            exit 0
+            ;;
+
+        "-version" | "--version" | "-v")
+            printf '%b\n' "$SIMGRID_VERSION"
+            exit 0
+            ;;
+
+        "--cfg="*|"--log="*)
+            for OPT in ${1#*=}
+            do
+                SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
+            done
+            shift 1
+            ;;
+        *)
+            break
+            ;;
+    esac
 done
 
-EXEC="$1"
+if [ -n "$WRAPPER" ]; then
+    EXEC="$WRAPPER $1"
+else
+    EXEC="$1"
+fi
 shift
 
+# steel --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
+
+
 ##-----------------------------------
 
-# test if we have something to execute, otherwise show usage and exit
-if [ -z ${EXEC} ]
-then
-    echo "Program is missing"
+
+if [ -z "${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 /.*<host.*?id="(.*?)".*?\/>.*/' ${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
+    UNROLLEDHOSTFILETMP=1
+    UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)"
+    perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
+    if [ ${HOSTFILETMP} = 1 ] ; then
+        rm ${HOSTFILE}
+        HOSTFILETMP=0
+    fi
+    HOSTFILE=$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]" $HOSTFILE`
+
+if [ -z "${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
 
 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
 if [ -z "${PLATFORM}" ]; then
-       PLATFORMTMP="$(mktemp tmpXXXXXX)"
+    PLATFORMTMP="$(mktemp tmpXXXXXX)"
 
-       cat > ${PLATFORMTMP} <<PLATFORMHEAD
+    cat > ${PLATFORMTMP} <<PLATFORMHEAD
 <?xml version='1.0'?>
 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
 <platform version="3">
 <AS id="AS0" routing="Full">
 PLATFORMHEAD
 
-for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
-  echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
-  echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
-  echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
-done
-
-for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
-  for (( j=${NUMPROCS}; $j ; j=$j-1 )) do
-    if [ $i -eq $j ]; then
-      echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
-    else
-      echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
-    fi
-  done
-done
-
-cat >> ${PLATFORMTMP} <<PLATFORMFOOT
+    i=${NUMPROCS}
+    while [ $i -gt 0 ]; do
+        echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
+        echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
+        echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
+        i=$((i - 1))
+    done
+
+    i=${NUMPROCS}
+    while [ $i -gt 0 ]; do
+        j=${NUMPROCS}
+        while [ $j -gt 0 ]; do
+            if [ $i -eq $j ]; then
+                echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
+            else
+                echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
+            fi
+            j=$((j - 1))
+        done
+        i=$((i - 1))
+    done
+
+    cat >> ${PLATFORMTMP} <<PLATFORMFOOT
 </AS>
 </platform>
 PLATFORMFOOT
 
 else
-       PLATFORMTMP=${PLATFORM}
+    PLATFORMTMP=${PLATFORM}
 fi
 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
 
@@ -189,8 +311,19 @@ APPLICATIONHEAD
 
 ##---- cache hostnames of hostfile---------------
 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
-       hostnames=(`cat ${HOSTFILE} | tr \\\n " "`)
-      NUMHOSTS=`cat ${HOSTFILE} | wc -l`
+    hostnames=$(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
 
 ##----------------------------------------------------------
@@ -203,38 +336,51 @@ fi
 HAVE_SEQ="`which seq 2>/dev/null`"
 
 if [ -n "${HAVE_SEQ}" ]; then
-       SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
+    SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
 else
-      cnt=0
-       while (( $cnt < ${NUMPROCS} )) ; do
-           SEQ="$SEQ $cnt"
-           cnt=$((cnt + 1));
-       done
+    cnt=0
+    while (( $cnt < ${NUMPROCS} )) ; do
+       SEQ="$SEQ $cnt"
+       cnt=$((cnt + 1));
+    done
 fi
 
 ##---- generate <process> tags------------------------------
 
 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
-
-  if [ -z "${hostnames[$j]}" ]; then
-       host="host"$(($j+1))
-  else
-       host="${hostnames[$j]}"
-  fi
-  echo "  <process host=\"${host}\" function=\"rank$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
-  echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
-  for ARG in $*; do
-    echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
-  done
-  echo "  </process>" >> ${APPLICATIONTMP}
+    if [ -n "${HOSTFILE}" ]; then
+        if [ ${NUMHOSTS} = 0 ] ; then
+          j= 0
+       else
+          j=$(( $i % ${NUMHOSTS} + 1 ))
+       fi
+    fi
+    ##---- optional display of ranks to process mapping
+    if [ -n "${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 "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
+    echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
+    echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
+    if [ "${EXTOPT}" = "smpi_replay" ]; then
+        if  [ ${NUMTRACES} -gt 1 ]; then
+            echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
+        else
+            echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
+        fi
+    else 
+        for ARG in $PROC_ARGS; do
+            echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
+        done
+    fi
+    echo "  </process>" >> ${APPLICATIONTMP}
 done
 
 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
@@ -244,35 +390,65 @@ APPLICATIONFOOT
 
 ##---------------------- 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:${TRACE_FILENAME} --cfg=tracing/smpi:1"
-
-  if [ -n "${TRACE_GROUPED}" ]; then
-    TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:1"
-  fi
-
-  if [ -n "${TRACE_RESOURCE}" ]; then
-    TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:1 --cfg=tracing/uncategorized:1"
-  fi
-
-  if [ -n "${TRACE_TRIVA}" ]; then
-    TRACEOPTIONS="${TRACEOPTIONS} --cfg=triva/categorized:smpi_cat.plist --cfg=triva/uncategorized:smpi_uncat.plist"
-  fi
+    #define trace filename
+    if [ -n "${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 "${TRACE_COMMENT}" ]; then
+        TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
+    fi
+
+    if [ -n "${TRACE_COMMENT_FILE}" ]; then
+        TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
+    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
+
+    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 ---------------------------------
 
 export SMPI_GLOBAL_SIZE=${NUMPROCS}
 if [ -n "${KEEP}" ] ; then
-  echo ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
+    echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
+    if [ ${HOSTFILETMP} = 1 ] ; then
+        echo "Generated hostfile ${HOSTFILE} keeped."
+    fi
+    if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
+        echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped." 
+    fi
 fi
-${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
+${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
+status=$?
 
 if [ -z "${KEEP}" ] ; then
-   if [ -z "${PLATFORM}" ]; then
+    if [ -z "${PLATFORM}" ]; then
        rm ${PLATFORMTMP}
-   fi
-   rm ${APPLICATIONTMP}
+    fi
+    if [ ${HOSTFILETMP} = 1 ] ; then
+        rm ${HOSTFILE}
+    fi
+    if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
+        rm ${UNROLLEDHOSTFILE}
+    fi
+    rm ${APPLICATIONTMP}
 fi
+
+exit $status