Re: [Starpu-devel] Bug trace

[Samuel Thibault <samuel.thibault@ens-lyon.org>]

Hello,

Berenger Bramas, le Wed 20 Mar 2013 17:36:27 +0100, a écrit :
> I got a segfault when I tried to convert a trace generated with 160 CPUs.
> The file is accessible from plafrim:
> /tmp/prof_file_bramas_0 on devel09

There was indeed a bug when there are more than 128 workers. Please try
the attached patch.

Samuel
Index: src/debug/traces/starpu_fxt.c
===================================================================
--- src/debug/traces/starpu_fxt.c       (révision 7187)
+++ src/debug/traces/starpu_fxt.c       (copie de travail)
@@ -80,7 +81,7 @@
 }
 
 static float last_codelet_start[STARPU_NMAXWORKERS];
-static char last_codelet_symbol[128][STARPU_NMAXWORKERS];
+static char last_codelet_symbol[STARPU_NMAXWORKERS][128];
 
 /* If more than a period of time has elapsed, we flush the profiling info,
  * otherwise they are accumulated everytime there is a new relevant event. */
@@ -353,7 +349,7 @@
        unsigned long has_name = ev->param[2];
        char *name = has_name?(char *)&ev->param[3]:"unknown";
 
-       snprintf(last_codelet_symbol[worker], 128, "%s", name);
+       snprintf(last_codelet_symbol[worker], 
sizeof(last_codelet_symbol[worker]), "%s", name);
 
        float start_codelet_time = get_event_time_stamp(ev, options);
        last_codelet_start[worker] = start_codelet_time;