Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
attach time-independent traces to processes when the executable is
[simgrid.git] / src / smpi / smpirun.in
1 #!/bin/bash
2
3 @CMAKE_SMPI_COMMAND@
4
5 DEFAULT_LOOPBACK_BANDWIDTH="498000000"
6 DEFAULT_LOOPBACK_LATENCY="0.000004"
7 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
8 DEFAULT_NETWORK_LATENCY="0.000005"
9 DEFAULT_NUMPROCS="4"
10 DEFAULT_POWER="100"
11
12 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
13 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
14 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
15 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
16 POWER="${DEFAULT_POWER}"
17
18 SIMOPTS="--cfg=maxmin/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
19
20 #usage to print the way this script should be called
21 function usage () {
22     echo "usage:"
23     echo "$0 [-keep-temps]"
24     echo "      [-np <numprocs>] -platform <xmldesc> -hostfile <hostfile> [-map] program [program-options]"
25     echo "      [-keep-temps]             # don't remove the generated files after execution"
26     echo "      [-map]                    # display the machine on which each process rank is mapped"
27     echo "      [-np <numprocs>]          # use that amount of processes from the hostfile."
28     echo "                                # By default, all processes of the hostfile are used."
29     echo "      [-trace]                  # activate tracing"
30     echo "      [-trace-grouped]          # group MPI processes by location"
31     echo "      [-trace-resource]         # trace resource utilization"
32     echo "      [-trace-triva]            # generate configuration for Triva's GraphView"
33     echo "      [-trace-file <tracefile>] # name of the tracefile (simgrid_smpi.trace)"
34     echo "or (deprecated usage):"
35     echo "$0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]"
36     echo
37 }
38
39 #check if we have at least one parameter
40 if [ $# -eq 0 ]
41 then
42     usage
43     exit
44 fi
45
46 while true; do
47   case "$1" in
48    "-np" | "-n")
49       NUMPROCS="$2"
50       shift 2
51     ;;
52    "-bandwidth")
53       NETWORK_BANDWIDTH="$2"
54       shift 2
55     ;;
56    "-latency")
57       NETWORK_LATENCY="$2"
58       shift 2
59     ;;
60    "-platform")
61         PLATFORM="$2"
62       if [ ! -f "${PLATFORM}" ]; then
63                 echo "[$0] ** error: the file '${PLATFORM}' does not exist. Aborting."
64                 exit 1
65       fi
66         shift 2
67     ;;
68    "-hostfile")
69         HOSTFILE="$2"
70       if [ ! -f "${HOSTFILE}" ]; then
71                 echo "[$0] ** error: the file '${HOSTFILE}' does not exist. Aborting."
72                 exit 1
73       fi
74         shift 2
75     ;;
76
77    "-map")
78         MAPOPT="true"
79       shift 1
80    ;;
81
82    "-trace")
83         TRACE_ACTIVE="true"
84           shift 1
85    ;;
86
87    "-trace-file")
88       TRACE_FILENAME="$2"
89       shift 2
90    ;;
91
92    "-trace-grouped")
93       TRACE_GROUPED="true"
94       shift 1
95    ;;
96
97    "-trace-resource")
98       TRACE_RESOURCE="true"
99       shift 1
100    ;;
101
102    "-trace-triva")
103         TRACE_TRIVA="true"
104           shift 1
105    ;;
106
107    "-keep-temps")
108         KEEP="true"
109       shift 1
110    ;;
111
112    "-help" | "--help" | "-h")
113       usage
114       exit
115    ;;
116
117    "--cfg="*|"--log="*)
118      for OPT in ${1#*=}
119      do
120        SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
121      done
122      shift 1
123    ;;
124     *)
125       break
126     ;;
127   esac
128 done
129
130 EXEC="$1"
131 shift
132
133 ##-----------------------------------
134
135 # Basic checks on the provided arguments
136 if [ -z "${EXEC}" ] ; then
137     echo "You must provide a program to execute."
138     usage
139     exit 1
140 fi
141
142 if [ -z "${HOSTFILE}" ] ; then
143     echo "No hostfile specified."
144     usage
145     exit 1
146 fi
147
148 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
149 hostfile_procs=`grep -c [a-zA-Z0-9] $HOSTFILE`
150
151 if [ -z "${NUMPROCS}" ] ; then
152     # Use the amount of processes in the hostfile as default value for the -np parameter
153     NUMPROCS=$hostfile_procs
154 fi
155
156
157 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
158     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
159 fi
160
161 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
162 if [ -z "${PLATFORM}" ]; then
163         PLATFORMTMP="$(mktemp tmpXXXXXX)"
164
165         cat > ${PLATFORMTMP} <<PLATFORMHEAD
166 <?xml version='1.0'?>
167 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
168 <platform version="3">
169 <AS id="AS0" routing="Full">
170 PLATFORMHEAD
171
172 for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
173   echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
174   echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
175   echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
176 done
177
178 for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
179   for (( j=${NUMPROCS}; $j ; j=$j-1 )) do
180     if [ $i -eq $j ]; then
181       echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
182     else
183       echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
184     fi
185   done
186 done
187
188 cat >> ${PLATFORMTMP} <<PLATFORMFOOT
189 </AS>
190 </platform>
191 PLATFORMFOOT
192
193 else
194         PLATFORMTMP=${PLATFORM}
195 fi
196 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
197
198 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
199 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
200 #APPLICATIONTMP="app.xml"
201
202
203 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
204 <?xml version='1.0'?>
205 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
206 <platform version="3">
207 APPLICATIONHEAD
208
209 ##---- cache hostnames of hostfile---------------
210 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
211         hostnames=(`cat ${HOSTFILE} | tr \\\n " "`)
212       NUMHOSTS=`cat ${HOSTFILE} | wc -l`
213 fi
214
215 if [ ${EXEC} = "./replay" ]; then 
216   APP_TRACES=$1;
217   if [ -n "${APP_TRACES}" ] && [ -f ${APP_TRACES} ]; then
218      hosttraces=(`cat ${APP_TRACES} | tr \\\n " "`)
219      NUMTRACES=`cat ${APP_TRACES} | wc -l`
220   fi
221 fi
222
223 ##----------------------------------------------------------
224 ##  generate application.xml with hostnames from hostfile:
225 ##  the name of host_i (1<=i<=p, where -np p) is the line i
226 ##  in hostfile (where -hostfile hostfile), or "host$i" if
227 ##  hostfile has less than i lines.
228 ##----------------------------------------------------------
229
230 HAVE_SEQ="`which seq 2>/dev/null`"
231
232 if [ -n "${HAVE_SEQ}" ]; then
233         SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
234 else
235       cnt=0
236         while (( $cnt < ${NUMPROCS} )) ; do
237             SEQ="$SEQ $cnt"
238             cnt=$((cnt + 1));
239         done
240 fi
241
242 ##---- generate <process> tags------------------------------
243
244 for i in ${SEQ}
245 do
246   if [ -n "${HOSTFILE}" ]; then
247         j=$(( $i % ${NUMHOSTS} ))
248   fi
249   ##---- optional display of ranks to process mapping
250   if [ -n "${MAPOPT}" ]; then
251         echo "[rank $i] -> ${hostnames[$j]}"
252   fi
253
254   if [ -z "${hostnames[$j]}" ]; then
255         host="host"$(($j+1))
256   else
257         host="${hostnames[$j]}"
258   fi
259   echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
260   echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
261   if [ ${EXEC} = "./replay" ]; then
262     if  [ ${NUMTRACES} -gt 1 ]; then
263       echo "    <argument value=\"${hosttraces[$j]}\"/>" >> ${APPLICATIONTMP}
264     else
265       echo "    <argument value=\"${hosttraces[0]}\"/>" >> ${APPLICATIONTMP}
266     fi
267   else 
268     for ARG in $*; do
269        echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
270     done
271   fi
272   echo "  </process>" >> ${APPLICATIONTMP}
273 done
274
275 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
276 </platform>
277 APPLICATIONFOOT
278 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
279
280 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
281 if [ -n "${TRACE_ACTIVE}" ]; then
282   #define trace filename
283   if [ -z "${TRACE_FILENAME}" ]; then
284     TRACE_FILENAME="smpi_simgrid.trace"
285   fi
286   TRACEOPTIONS="--cfg=tracing:1 --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:1"
287
288   if [ -n "${TRACE_GROUPED}" ]; then
289     TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:1"
290   fi
291
292   if [ -n "${TRACE_RESOURCE}" ]; then
293     TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:1 --cfg=tracing/uncategorized:1"
294   fi
295
296   if [ -n "${TRACE_TRIVA}" ]; then
297     TRACEOPTIONS="${TRACEOPTIONS} --cfg=triva/categorized:smpi_cat.plist --cfg=triva/uncategorized:smpi_uncat.plist"
298   fi
299 fi
300 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
301
302 export SMPI_GLOBAL_SIZE=${NUMPROCS}
303 if [ -n "${KEEP}" ] ; then
304   echo ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
305 fi
306 ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
307 status=$?
308
309 if [ -z "${KEEP}" ] ; then
310    if [ -z "${PLATFORM}" ]; then
311         rm ${PLATFORMTMP}
312    fi
313    rm ${APPLICATIONTMP}
314 fi
315
316 exit $status