Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
smpirun: cleanup files and children before dying
[simgrid.git] / src / smpi / smpirun.in
1 #! /bin/sh
2
3 # Copyright (c) 2007-2015. The SimGrid Team.
4 # All rights reserved.
5
6 # This program is free software; you can redistribute it and/or modify it
7 # under the terms of the license (GNU LGPL) which comes with this package.
8
9 @CMAKE_SMPI_COMMAND@
10
11 SIMGRID_VERSION="@SIMGRID_VERSION_STRING@"
12 SIMGRID_GITHASH="@SIMGRID_GITHASH@"
13
14 DEFAULT_LOOPBACK_BANDWIDTH="498000000"
15 DEFAULT_LOOPBACK_LATENCY="0.000004"
16 DEFAULT_NETWORK_BANDWIDTH="$((26 * 1024 * 1024))"
17 DEFAULT_NETWORK_LATENCY="0.000005"
18 DEFAULT_NUMPROCS="4"
19 DEFAULT_POWER="100"
20
21 LOOPBACK_BANDWIDTH="${DEFAULT_LOOPBACK_BANDWIDTH}"
22 LOOPBACK_LATENCY="${DEFAULT_LOOPBACK_LATENCY}"
23 NETWORK_BANDWIDTH="${DEFAULT_NETWORK_BANDWIDTH}"
24 NETWORK_LATENCY="${DEFAULT_NETWORK_LATENCY}"
25 POWER="${DEFAULT_POWER}"
26
27 SIMOPTS="--cfg=surf/precision:1e-9 --cfg=network/model:SMPI --cfg=network/TCP_gamma:4194304"
28
29 #usage to print the way this script should be called
30 usage () {
31     cat <<EOF
32 Usage: $0 [OPTIONS] -platform <xmldesc> -hostfile <hostfile> program [program-options]
33 Options:
34   -keep-temps                # don't remove the generated files after execution
35   -wrapper <command>         # use command to run the program (e.g. "valgrind")
36   -map                       # display the machine on which each process rank is mapped
37   -np <numprocs>             # use that amount of processes from the hostfile.
38                              # By default, all processes of the hostfile are used.
39   -trace-ti                  # activate time independant tracing (for replay, default in smpi_simgrid.txt)
40   -trace                     # activate tracing (Paje, default in smpi_simgrid.trace)
41   -trace-comment <comment>   # put a comment on the top of the trace file
42   -trace-comment-file <file> # put file contents on the top of the trace file as comment
43   -trace-grouped             # group MPI processes by location
44   -trace-resource            # trace resource utilization
45   -trace-viva                # generate configuration for Viva's GraphView
46   -trace-file <tracefile>    # name of the tracefile (simgrid_smpi.trace)
47   -ext <value>               # additional parameter (reserved)
48   
49   -version                   # Displays the SimGrid version (human readable)
50   -git-version               # Displays the git hash of SimGrid
51
52 or (deprecated usage):
53   $0 [-keep-temps] [-np <numprocs>] [-bandwidth <bytes/sec>] [-latency <secs>] program [program-options]
54
55 EOF
56 }
57
58 #check if we have at least one parameter
59 if [ $# -eq 0 ]
60 then
61     usage
62     exit
63 fi
64
65 EXTOPT=""
66 WRAPPER=""
67 HOSTFILE=""
68 HOSTFILETMP=0
69
70 unset pid
71
72 smpirun_cleanup()
73 {
74   if [ -z "${KEEP}" ] ; then
75       if [ -z "${PLATFORM}" -a -n "$PLATFORMTMP" ]; then
76         rm -f ${PLATFORMTMP}
77         PLATFORMTMP=""
78       fi
79       if [ ${HOSTFILETMP} = 1 -a -n "$HOSTFILE" ] ; then
80           rm -f ${HOSTFILE}
81           HOSTFILE=""
82       fi
83       if [ ${UNROLLEDHOSTFILETMP} = 1 -a -n "$UNROLLEDHOSTFILE" ] ; then
84           rm -f ${UNROLLEDHOSTFILE}
85           UNROLLEDHOSTFILE=""
86       fi
87       if [ -n ${APPLICATIONTMP} ]; then
88         rm -f ${APPLICATIONTMP}
89         APPLICATIONTMP=""
90       fi
91   fi
92 }
93
94 smpirun_trap() {
95   smpirun_cleanup
96   if ! [ -z "$pid" ]; then
97     kill -TERM $pid
98   fi
99   unset pid
100   trap - HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS
101   kill -TERM $$
102   exit 128
103 }
104
105 trap smpirun_trap HUP INT QUIT ILL ABRT SEGV FPE ALRM TERM USR1 USR2 BUS
106
107 while true; do
108     case "$1" in
109         "-np" | "-n")
110             NUMPROCS="$2"
111             shift 2
112             ;;
113         "-bandwidth")
114             NETWORK_BANDWIDTH="$2"
115             shift 2
116             ;;
117         "-latency")
118             NETWORK_LATENCY="$2"
119             shift 2
120             ;;
121         "-platform")
122             PLATFORM="$2"
123             if [ ! -f "${PLATFORM}" ]; then
124                 echo "[`basename $0`] ** error: the file '${PLATFORM}' does not exist. Aborting."
125                 exit 1
126             fi
127             shift 2
128             ;;
129         "-hostfile")
130             HOSTFILE="$2"
131             if [ ! -f "${HOSTFILE}" ]; then
132                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
133                 exit 1
134             fi
135             shift 2
136             ;;
137
138         "-machinefile")
139             HOSTFILE="$2"
140             if [ ! -f "${HOSTFILE}" ]; then
141                 echo "[`basename $0`] ** error: the file '${HOSTFILE}' does not exist. Aborting."
142                 exit 1
143             fi
144             shift 2
145             ;;
146
147         "-ext")
148             EXTOPT="$2"
149             shift 2
150             ;;
151
152         "-map")
153             MAPOPT="true"
154             shift 1
155             ;;
156
157         "-trace")
158             TRACE_ACTIVE="true"
159             shift 1
160             ;;
161
162         "-trace-ti")
163             TRACE_ACTIVE="true"
164             TRACE_TI_ACTIVE="true"
165             shift 1
166             ;;
167
168         "-trace-comment")
169             TRACE_COMMENT="$2"
170             shift 2
171             ;;
172
173         "-trace-comment-file")
174             TRACE_COMMENT_FILE="$2"
175             shift 2
176             ;;
177
178         "-trace-file")
179             TRACE_FILENAME="$2"
180             shift 2
181             ;;
182
183         "-trace-grouped")
184             TRACE_GROUPED="true"
185             shift 1
186             ;;
187
188         "-trace-resource")
189             TRACE_RESOURCE="true"
190             shift 1
191             ;;
192
193         "-trace-viva")
194             TRACE_VIVA="true"
195             shift 1
196             ;;
197
198         "-keep-temps")
199             KEEP="true"
200             shift 1
201             ;;
202
203         "-wrapper")
204             WRAPPER="$2"
205             shift 2
206             ;;
207
208         "-help" | "--help" | "-h")
209             usage
210             exit 0
211             ;;
212
213         "-version" | "--version" | "-v")
214             printf '%b\n' "$SIMGRID_VERSION"
215             exit 0
216             ;;
217             
218         "-git-version" | "--git-version")
219             printf '%b\n' "$SIMGRID_GITHASH"
220             exit 0
221             ;;
222
223         "--cfg="*|"--log="*)
224             for OPT in ${1#*=}
225             do
226                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
227             done
228             shift 1
229             ;;
230         *)
231             break
232             ;;
233     esac
234 done
235
236 if [ -n "$WRAPPER" ]; then
237     EXEC="$WRAPPER $1"
238 else
239     EXEC="$1"
240 fi
241 shift
242
243 # steel --cfg and --logs options
244 while [ $# -gt 0 ]; do
245     case "$1" in
246         "--cfg="*|"--log="*)
247             for OPT in ${1#*=}
248             do
249                 SIMOPTS="$SIMOPTS ${1%%=*}=$OPT"
250             done
251             shift 1
252             ;;
253         *)
254             PROC_ARGS="${PROC_ARGS:+$PROC_ARGS }$1"
255             shift      
256             ;;
257     esac
258 done
259
260
261 ##-----------------------------------
262
263
264 if [ -z "${HOSTFILE}" ] && [ -z "${PLATFORM}" ] ; then
265     echo "No hostfile nor platform specified."
266     usage
267     exit 1
268 fi
269
270 if [ -z "${HOSTFILE}" ] ; then
271     HOSTFILETMP=1
272     HOSTFILE="$(mktemp tmphostXXXXXX)"
273     perl -ne 'print "$1\n" if /.*<host.*?id="(.*?)".*?\/>.*/' ${PLATFORM} > ${HOSTFILE}
274     perl -ne 'if (/.*<cluster.*?prefix="(.*?)".*?radical="(.*?)".*?suffix="(.*?)".*/) { 
275                 my ($pre,$rad,$post)=($1,$2,$3); 
276                 for my $elm (split(",",$rad)) { 
277                   if ($elm=~/^([^-]*?)-([^-]*)$/) { 
278                      for (my $i=$1; $i<=$2;$i++) { 
279                         print "$pre$i$post\n"; 
280                      }
281                   } else {
282                      print "$pre$elm$post\n";
283                   }
284                 }
285               } elsif (/<cluster/) {
286                 die ("Unparsable cluster tag. Either provide an hostfile yourself or give the attributes prefix, radical and suffix in that order on the <cluster line");
287               }' ${PLATFORM} >> ${HOSTFILE}
288 fi
289 UNROLLEDHOSTFILETMP=0
290
291 #parse if our lines are terminated by :num_process
292 multiple_processes=`grep -c ":" $HOSTFILE`
293 if [ "${multiple_processes}" -gt 0 ] ; then
294     UNROLLEDHOSTFILETMP=1
295     UNROLLEDHOSTFILE="$(mktemp tmphostXXXXXX)"
296     perl -ne ' do{ for ( 1 .. $2 ) { print "$1\n" } } if /(.*?):(\d+).*/'  ${HOSTFILE}  > ${UNROLLEDHOSTFILE}
297     if [ ${HOSTFILETMP} = 1 ] ; then
298         rm ${HOSTFILE}
299         HOSTFILETMP=0
300     fi
301     HOSTFILE=$UNROLLEDHOSTFILE
302 fi
303
304
305 # Don't use wc -l to compute it to avoid issues with trailing \n at EOF
306 hostfile_procs=`grep -c "[a-zA-Z0-9]" $HOSTFILE`
307 if [ ${hostfile_procs} = 0 ] ; then
308    echo "[`basename $0`] ** error: the hostfile '${HOSTFILE}' is empty. Aborting." >&2
309    exit 1
310 fi
311
312 if [ -z "${NUMPROCS}" ] ; then
313     # Use the amount of processes in the hostfile as default value for the -np parameter
314     NUMPROCS=$hostfile_procs
315 fi
316
317
318 if [ ${NUMPROCS} -gt ${hostfile_procs} ] ; then
319     echo "You requested to use ${NUMPROCS} processes, but there is only ${hostfile_procs} processes in your hostfile..." >&2
320 fi
321
322 ##-------------------------------- DEFAULT or SPECIFIED PLATFORM --------------------------------------
323 if [ -z "${PLATFORM}" ]; then
324     PLATFORMTMP="$(mktemp tmpXXXXXX)"
325
326     cat > ${PLATFORMTMP} <<PLATFORMHEAD
327 <?xml version='1.0'?>
328 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
329 <platform version="3">
330 <AS id="AS0" routing="Full">
331 PLATFORMHEAD
332
333     i=${NUMPROCS}
334     while [ $i -gt 0 ]; do
335         echo "  <host id=\"host$i\" power=\"${POWER}\"/>" >> ${PLATFORMTMP}
336         echo "  <link id=\"loop$i\" bandwidth=\"${LOOPBACK_BANDWIDTH}\" latency=\"${LOOPBACK_LATENCY}\"/>" >> ${PLATFORMTMP}
337         echo "  <link id=\"link$i\" bandwidth=\"${NETWORK_BANDWIDTH}\" latency=\"${NETWORK_LATENCY}\"/>" >> ${PLATFORMTMP}
338         i=$((i - 1))
339     done
340
341     i=${NUMPROCS}
342     while [ $i -gt 0 ]; do
343         j=${NUMPROCS}
344         while [ $j -gt 0 ]; do
345             if [ $i -eq $j ]; then
346                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"loop$i\"/></route>" >> ${PLATFORMTMP}
347             else
348                 echo "  <route src=\"host$i\" dst=\"host$j\"><link_ctn id=\"link$i\"/><link_ctn id=\"link$j\"/></route>" >> ${PLATFORMTMP}
349             fi
350             j=$((j - 1))
351         done
352         i=$((i - 1))
353     done
354
355     cat >> ${PLATFORMTMP} <<PLATFORMFOOT
356 </AS>
357 </platform>
358 PLATFORMFOOT
359
360 else
361     PLATFORMTMP=${PLATFORM}
362 fi
363 ##-------------------------------- end DEFAULT or SPECIFIED PLATFORM --------------------------------------
364
365 ##-------------------------------- DEFAULT APPLICATION --------------------------------------
366 APPLICATIONTMP="$(mktemp tmpXXXXXX)"
367 #APPLICATIONTMP="app.xml"
368
369
370 cat > ${APPLICATIONTMP} <<APPLICATIONHEAD
371 <?xml version='1.0'?>
372 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
373 <platform version="3">
374 APPLICATIONHEAD
375
376 ##---- cache hostnames of hostfile---------------
377 if [ -n "${HOSTFILE}" ] && [ -f ${HOSTFILE} ]; then
378     hostnames=$(cat ${HOSTFILE} | tr '\n\r' '  ')
379     NUMHOSTS=$(cat ${HOSTFILE} | wc -l)
380 fi
381
382 if [ "${EXTOPT}" = "smpi_replay" ]; then
383     APP_TRACES=$PROC_ARGS
384     if [ -n "${APP_TRACES}" ] && [ -f "${APP_TRACES}" ]; then
385         hosttraces=$(cat ${APP_TRACES} | tr '\n\r' '  ' )
386         NUMTRACES=$(cat ${APP_TRACES} | wc -l)
387     else
388         printf "File not found: %s\n", "${APP_TRACES:-\${APP_TRACES\}}" >&2
389         exit 1
390     fi
391 fi
392
393 ##----------------------------------------------------------
394 ##  generate application.xml with hostnames from hostfile:
395 ##  the name of host_i (1<=i<=p, where -np p) is the line i
396 ##  in hostfile (where -hostfile hostfile), or "host$i" if
397 ##  hostfile has less than i lines.
398 ##----------------------------------------------------------
399
400 HAVE_SEQ="`which seq 2>/dev/null`"
401
402 if [ -n "${HAVE_SEQ}" ]; then
403     SEQ=`${HAVE_SEQ} 0 $((${NUMPROCS}-1))`
404 else
405     cnt=0
406     while [ $cnt -lt ${NUMPROCS} ] ; do
407         SEQ="$SEQ $cnt"
408         cnt=$((cnt + 1));
409     done
410 fi
411
412 ##---- generate <process> tags------------------------------
413
414 for i in ${SEQ}
415 do
416     if [ -n "${HOSTFILE}" ]; then
417         j=$(( $i % ${NUMHOSTS} + 1 ))
418     fi
419     ##---- optional display of ranks to process mapping
420     if [ -n "${MAPOPT}" ]; then
421         echo "[rank $i] -> $(echo $hostnames|cut -d' ' -f$j)"
422     fi
423
424     if [ -z "$(echo $hostnames|cut -d' ' -f$j)" ]; then
425         host="host"$($j)
426     else
427         host="$(echo $hostnames|cut -d' ' -f$j)"
428     fi
429     echo "  <process host=\"${host}\" function=\"$i\"> <!-- function name used only for logging -->" >> ${APPLICATIONTMP}
430     echo "    <argument value=\"1\"/> <!-- instance -->" >> ${APPLICATIONTMP}
431     echo "    <argument value=\"$i\"/> <!-- rank -->" >> ${APPLICATIONTMP}
432     if [ "${EXTOPT}" = "smpi_replay" ]; then
433         if  [ ${NUMTRACES} -gt 1 ]; then
434             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f$j)\"/>" >> ${APPLICATIONTMP}
435         else
436             echo "    <argument value=\"$(echo $hosttraces|cut -d' ' -f1)\"/>" >> ${APPLICATIONTMP}
437         fi
438     else 
439         for ARG in $PROC_ARGS; do
440             echo "    <argument value=\"${ARG}\"/>" >> ${APPLICATIONTMP}
441         done
442     fi
443     echo "  </process>" >> ${APPLICATIONTMP}
444 done
445
446 cat >> ${APPLICATIONTMP} <<APPLICATIONFOOT
447 </platform>
448 APPLICATIONFOOT
449 ##-------------------------------- end DEFAULT APPLICATION --------------------------------------
450
451 ##---------------------- SMPI TRACING OPTIONS ---------------------------------
452 if [ -n "${TRACE_ACTIVE}" ]; then
453     #define trace filename
454     if [ -n "${TRACE_TI_ACTIVE}" ]; then
455         if [ -z "${TRACE_FILENAME}" ]; then
456             TRACE_FILENAME="smpi_simgrid.txt"
457         fi
458         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes --cfg=tracing/smpi/format:TI --cfg=tracing/smpi/computing:yes"
459     else
460         if [ -z "${TRACE_FILENAME}" ]; then
461             TRACE_FILENAME="smpi_simgrid.trace"
462         fi
463         TRACEOPTIONS="--cfg=tracing:yes --cfg=tracing/filename:${TRACE_FILENAME} --cfg=tracing/smpi:yes"
464     fi
465
466     if [ -n "${TRACE_COMMENT}" ]; then
467         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment:${TRACE_COMMENT}"
468     fi
469
470     if [ -n "${TRACE_COMMENT_FILE}" ]; then
471         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/comment_file:${TRACE_COMMENT_FILE}"
472     fi
473
474     if [ -n "${TRACE_GROUPED}" ]; then
475         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/smpi/group:yes"
476     fi
477
478     if [ -n "${TRACE_RESOURCE}" ]; then
479         TRACEOPTIONS="${TRACEOPTIONS} --cfg=tracing/categorized:yes --cfg=tracing/uncategorized:yes"
480     fi
481
482     if [ -n "${TRACE_VIVA}" ]; then
483         TRACEOPTIONS="${TRACEOPTIONS} --cfg=viva/categorized:smpi_cat.plist --cfg=viva/uncategorized:smpi_uncat.plist"
484     fi
485 fi
486 ##---------------------- end SMPI TRACING OPTIONS ---------------------------------
487
488 export SMPI_GLOBAL_SIZE=${NUMPROCS}
489 if [ -n "${KEEP}" ] ; then
490     echo ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP}
491     if [ ${HOSTFILETMP} = 1 ] ; then
492         echo "Generated hostfile ${HOSTFILE} keeped."
493     fi
494     if [ ${UNROLLEDHOSTFILETMP} = 1 ] ; then
495         echo "Generated unrolled hostfile ${UNROLLEDHOSTFILE} keeped." 
496     fi
497 fi
498 ${EXEC} ${TRACEOPTIONS} ${SIMOPTS} ${PLATFORMTMP} ${APPLICATIONTMP} &
499 pid=$!
500 wait $pid
501 status=$?
502 pid=""
503
504 smpirun_cleanup
505
506 exit $status