Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
b4ded8a4097a0aa8df99ef42f817cb769eeef910
[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
88    "-trace-file")
89       TRACE_FILENAME="$2"
90       shift 2
91    ;;
92
93    "-trace-grouped")
94       TRACE_GROUPED="true"
95       shift 1
96    ;;
97
98    "-trace-resource")
99       TRACE_RESOURCE="true"
100       shift 1
101    ;;
102
103    "-trace-triva")
104         TRACE_TRIVA="true"
105           shift 1
106    ;;
107
108    "-keep-temps")
109         KEEP="true"
110       shift 1
111    ;;
112
113    "-help" | "--help" | "-h")
114       usage
115       exit
116    ;;
117
118    "--cfg="*|"--log="*)
119      for OPT in ${1#*=}
120      do
121        SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
122      done
123      shift 1
124    ;;
125     *)
126       break
127     ;;
128   esac
129 done
130
131 EXEC="$1"
132 shift
133
134 ##-----------------------------------
135
136 # Basic checks on the provided arguments
137 if [ -z "${EXEC}" ] ; then
138     echo "You must provide a program to execute."
139     usage
140     exit 1
141 fi
142
143 if [ -z "${HOSTFILE}" ] ; then
144     echo "No hostfile specified."
145     usage
146     exit 1
147 fi
148
149 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
150 hostfile_procs=`grep -c [a-zA-Z0-9] $HOSTFILE`
151
152 if [ -z "${NUMPROCS}" ] ; then
153     # Use the amount of processes in the hostfile as default value for the -np parameter
154     NUMPROCS=$hostfile_procs
155 fi
156
157
158 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
159     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
160 fi
161
162 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
163 if [ -z "${PLATFORM}" ]; then
164         PLATFORMTMP="$(mktemp tmpXXXXXX)"
165
166         cat > ${PLATFORMTMP} <<PLATFORMHEAD
167 <?xml version='1.0'?>
168 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
169 <platform version="3">
170 <AS id="AS0" routing="Full">
171 PLATFORMHEAD
172
173 for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
174   echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
175   echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
176   echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
177 done
178
179 for (( i=${NUMPROCS}; $i ; i=$i-1 )) do
180   for (( j=${NUMPROCS}; $j ; j=$j-1 )) do
181     if [ $i -eq $j ]; then
182       echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
183     else
184       echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
185     fi
186   done
187 done
188
189 cat >> ${PLATFORMTMP} <<PLATFORMFOOT
190 </AS>
191 </platform>
192 PLATFORMFOOT
193
194 else
195         PLATFORMTMP=${PLATFORM}
196 fi
197 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
198
199 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
200 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
201 #APPLICATIONTMP="app.xml"
202
203
204 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
205 <?xml version='1.0'?>
206 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
207 <platform version="3">
208 APPLICATIONHEAD
209
210 ##---- cache hostnames of hostfile---------------
211 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
212         hostnames=(`cat ${HOSTFILE} | tr \\\n " "`)
213       NUMHOSTS=`cat ${HOSTFILE} | wc -l`
214 fi
215
216 if [ ${EXEC} = "./replay" ]; then 
217   APP_TRACES=$1;
218   if [ -n "${APP_TRACES}" ] && [ -f ${APP_TRACES} ]; then
219      hosttraces=(`cat ${APP_TRACES} | tr \\\n " "`)
220      NUMTRACES=`cat ${APP_TRACES} | wc -l`
221   fi
222 fi
223
224 ##----------------------------------------------------------
225 ##  generate application.xml with hostnames from hostfile:
226 ##  the name of host_i (1<=i<=p, where -np p) is the line i
227 ##  in hostfile (where -hostfile hostfile), or "host$i" if
228 ##  hostfile has less than i lines.
229 ##----------------------------------------------------------
230
231 HAVE_SEQ="`which seq 2>/dev/null`"
232
233 if [ -n "${HAVE_SEQ}" ]; then
234         SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
235 else
236       cnt=0
237         while (( $cnt < ${NUMPROCS} )) ; do
238             SEQ="$SEQ $cnt"
239             cnt=$((cnt + 1));
240         done
241 fi
242
243 ##---- generate <process> tags------------------------------
244
245 for i in ${SEQ}
246 do
247   if [ -n "${HOSTFILE}" ]; then
248         j=$(( $i % ${NUMHOSTS} ))
249   fi
250   ##---- optional display of ranks to process mapping
251   if [ -n "${MAPOPT}" ]; then
252         echo "[rank $i] -> ${hostnames[$j]}"
253   fi
254
255   if [ -z "${hostnames[$j]}" ]; then
256         host="host"$(($j+1))
257   else
258         host="${hostnames[$j]}"
259   fi
260   echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
261   echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
262   if [ ${EXEC} = "./replay" ]; then
263     if  [ ${NUMTRACES} -gt 1 ]; then
264       echo "    <argument value=\"${hosttraces[$j]}\"/>" >> ${APPLICATIONTMP}
265     else
266       echo "    <argument value=\"${hosttraces[0]}\"/>" >> ${APPLICATIONTMP}
267     fi
268   else 
269     for ARG in $*; do
270        echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
271     done
272   fi
273   echo "  </process>" >> ${APPLICATIONTMP}
274 done
275
276 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
277 </platform>
278 APPLICATIONFOOT
279 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
280
281 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
282 if [ -n "${TRACE_ACTIVE}" ]; then
283   #define trace filename
284   if [ -z "${TRACE_FILENAME}" ]; then
285     TRACE_FILENAME="smpi_simgrid.trace"
286   fi
287   TRACEOPTIONS="--cfg=tracing:1 --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:1"
288
289   if [ -n "${TRACE_GROUPED}" ]; then
290     TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:1"
291   fi
292
293   if [ -n "${TRACE_RESOURCE}" ]; then
294     TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:1 --cfg=tracing/uncategorized:1"
295   fi
296
297   if [ -n "${TRACE_TRIVA}" ]; then
298     TRACEOPTIONS="${TRACEOPTIONS} --cfg=triva/categorized:smpi_cat.plist --cfg=triva/uncategorized:smpi_uncat.plist"
299   fi
300 fi
301 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
302
303 export SMPI_GLOBAL_SIZE=${NUMPROCS}
304 if [ -n "${KEEP}" ] ; then
305   echo ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
306 fi
307 ${EXEC} ${SIMOPTS} ${TRACEOPTIONS} ${PLATFORMTMP} ${APPLICATIONTMP}
308 status=$?
309
310 if [ -z "${KEEP}" ] ; then
311    if [ -z "${PLATFORM}" ]; then
312         rm ${PLATFORMTMP}
313    fi
314    rm ${APPLICATIONTMP}
315 fi
316
317 exit $status