VT enhancements:

- extendet support for BlueGene/P:
	- building shared VT libraries
	- tracing 3rd-party libraries (e.g. libc I/O)
	- tracing multi-threaded applications 
VT configure fixes:
- fixed detection on CTool for 3rd-party library tracing
VT fixes:
- reduced memory overhead by using the trace buffer for string/array elements of some records
- do not shutdown call-stack if max. number of buffer flushes reached, because the additional function leaves suggest a wrong application flow
- vtunify-mpi:
	- fixed conversion of VTUnify_MPI_Aint arrays 
- vtwrapper:
	- if an OPARI modified object file (*.mod.o) cannot be renamed, abort only if the compiler wrapper runs in "only-compile" mode (-c) 
OTF fixes:
- otfinfo:
	- fixed and enhanced calculation of trace file size
	- changed unit of timer resolution (s -> Hz) 
- otfprofile:
	- fixed progress
	- kill '_' and '\' in process names to make LaTex happier

This commit was SVN r22963.

ompi/contrib/vt/vt/ChangeLog

@@ -3,6 +3,8 @@
 	  (see extlib/otf/ChangeLog)
 	- added support for tracing functions in shared libraries instrumented
 	  by the GNU, Intel, PathScale, or PGI 9 compiler
+	- extended support for BlueGene/P in terms of tracing libraries
+	  (e.g. libc I/O) and tracing of multi-threaded applications
 	- added support for PAPI-C counters which belong to different components
 	- added environment variable VT_METRICS_SEP to specify a string which
 	  separate the counter names in VT_METRICS
@@ -20,13 +22,15 @@
 	- use more portable 'nm' command instead of the BFD library to collect
 	  symbol information for instrumentation with the GNU, Intel, PathScale,
 	  and PGI 9 compiler
-	- fixed stack shutdown if maximum number of buffer flushes was reached
+	- do not shutdown call stack if maximum number of buffer flushes was
+	  reached, because the additional leaves suggest a wrong application
+	  flow
 	- fixed potential stack underflow in vtfilter which might be cause a
 	  segmentation fault
 	- fixed detection of unique node id on MacOS platforms
 	- fixed support for LAM/MPI
 	- fixed potential segmentation fault in vtunify-mpi which might occur
-	  on 32bit platforms using MPICH2
+	  on 32bit platforms
 	- patched libtool to avoid a bug in detection of the PGI 10 compilers
 
 5.8openmpi

ompi/contrib/vt/vt/config/defaults/bgl

@@ -8,8 +8,7 @@ CC_FOR_BUILD="xlc_r"
 CXX_FOR_BUILD="xlC_r"
 enable_shared="no"
 enable_threads="no"
-enable_libctrace="no"
+enable_libtrace="no"
-enable_iotrace="no"
 enable_cpuidtrace="no"
 with_cross_prefix="blrts_"
 with_mpibgl="yes"

ompi/contrib/vt/vt/config/defaults/bgp

@@ -7,9 +7,6 @@ MPICXX="mpixlcxx_r"
 CC_FOR_BUILD="xlc_r"
 CXX_FOR_BUILD="xlC_r"
 enable_shared="no"
-enable_threads="no"
-enable_libctrace="no"
-enable_iotrace="no"
 enable_cpuidtrace="no"
 with_cross_prefix="bg"
 with_mpibgp="yes"

ompi/contrib/vt/vt/config/defaults/necsx

@@ -17,8 +17,7 @@ OPENMP_CFLAGS="-Popenmp"
 PTHREAD_LIBS="-lpthread"
 enable_shared="no"
 enable_memtrace="no"
-enable_libctrace="no"
+enable_libtrace="no"
-enable_iotrace="no"
 enable_cpuidtrace="no"
 enable_mpi2_io="no"
 enable_metrics="necsx"

ompi/contrib/vt/vt/config/m4/acinclude.compwrap.m4

@@ -24,6 +24,7 @@ AC_DEFUN([ACVT_COMPWRAP],
 	VT_WRAPPER_AVAIL_INST="manual"
 	VT_WRAPPER_DEFAULT_INST=manual
 
+	AC_REQUIRE([ACVT_PLATFORM])
 	AC_REQUIRE([ACVT_COMPINST])
 	AC_REQUIRE([ACVT_DYNINST])
 	AC_REQUIRE([ACVT_THREADS])
@@ -78,6 +79,11 @@ AC_DEFUN([ACVT_COMPWRAP],
 		[extra flags to add to LIBS when using wrapper compilers]),
 	[VT_WRAPPER_EXTRA_LIBS=$withval])
 
+	AS_IF([test "$PLATFORM" = "bgp" -a x"$enable_shared" = "xyes"],
+	[
+		VT_WRAPPER_EXTRA_LDFLAGS="$VT_WRAPPER_EXTRA_LDFLAGS -Wl,-dy"
+	])
+
 	VT_WRAPPER_VTLIB="-lvt"
 	VT_WRAPPER_VTMPILIB="-lvt-mpi"
 	VT_WRAPPER_VTMTLIB="-lvt-mt"

ompi/contrib/vt/vt/config/m4/acinclude.ctool.m4

@@ -37,23 +37,23 @@ AC_DEFUN([ACVT_CTOOL],
 	sav_LDFLAGS=$LDFLAGS; LDFLAGS=$LDFLAGS_FOR_BUILD
 	sav_LIBS=$LIBS; LIBS=$LIBS_FOR_BUILD
 
-	sav_CPPFLAGS=$CPPFLAGS
+	sav2_CPPFLAGS=$CPPFLAGS
 	CPPFLAGS="$CPPFLAGS $CTOOLINCDIR"
         AC_CHECK_HEADER([ctool/ctool.h], [],
 	[
 		AC_MSG_NOTICE([error: no ctool/ctool.h found; check path for CTool package first...])
 		ctool_error="yes"
 	])
-	CPPFLAGS=$sav_CPPFLAGS
+	CPPFLAGS=$sav2_CPPFLAGS
 
 	AS_IF([test x"$CTOOLLIB" = x -a x"$ctool_error" = "xno"],
 	[
-		sav_LIBS=$LIBS
+		sav2_LIBS=$LIBS
 		LIBS="$LIBS $CTOOLLIBDIR -lctool"
 		AC_MSG_CHECKING([whether linking with -lctool works])
 		AC_TRY_LINK([],[],
 		[AC_MSG_RESULT([yes]); CTOOLLIB=-lctool],[AC_MSG_RESULT([no])])
-		LIBS=$sav_LIBS
+		LIBS=$sav2_LIBS
 	])
 
 	CXX=$sav_CXX

ompi/contrib/vt/vt/config/m4/acinclude.dl.m4

@@ -33,11 +33,15 @@ AC_DEFUN([ACVT_DL],
 		AC_HELP_STRING([--with-dl-lib=DLLIB], [use given libdl lib, default: -ldl]),
 	[DLLIB="$withval"])
 
-	AS_IF([test "$PLATFORM" = "bgl" -o "$PLATFORM" = "bgp" -o "$PLATFORM" = "crayxt"],
+	AS_IF([test "$PLATFORM" = "bgl" -o "$PLATFORM" = "crayxt"],
 	[
 		AC_MSG_NOTICE([error: dynamic linking library (libdl) isn't suitable on this platform])
 		dl_error="yes"
 	])
+	AS_IF([test "$PLATFORM" = "bgp"],
+	[
+		ac_cv_have_decl_RTLD_NEXT="no"
+	])
 
 	AS_IF([test x"$dl_error" = "xno"],
 	[

ompi/contrib/vt/vt/config/m4/acinclude.java.m4

@@ -5,16 +5,27 @@ AC_DEFUN([ACVT_JAVA],
 	force_java="no"
 	have_java="no"
 
+	AC_REQUIRE([ACVT_PLATFORM])
+
 	AC_ARG_ENABLE(java, 
 		AC_HELP_STRING([--enable-java],
 		[enable Java support, default: enable if JVMTI found by configure]),
 	[AS_IF([test x"$enableval" = "xyes"], [force_java="yes"], [check_java="no"])])
 
 	AS_IF([test x"$check_java" = "xyes"],
+	[
+		AS_IF([test "$PLATFORM" = "bgp"],
+		[
+			AC_MSG_NOTICE([error: Java tracing not supported on this platform])
+			java_error="yes"
+		])
+
+		AS_IF([test x"$java_error" = "xno"],
 		[
 			AC_MSG_CHECKING([whether we can build shared libraries])
 			AS_IF([test x"$enable_shared" = "xyes"],
 			[AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no]); java_error="yes"])
+		])
 
 		AS_IF([test x"$java_error" = "xno"],
 		[

ompi/contrib/vt/vt/config/m4/acinclude.libc.m4

@@ -4,6 +4,8 @@ AC_DEFUN([ACVT_LIBC],
 
 	libc_pathname=
 
+	AC_REQUIRE([ACVT_PLATFORM])
+
 	AC_ARG_WITH(libc,
 		AC_HELP_STRING([--with-libc=NAME],
 		[give the pathname for LIBC, default: automatically by configure]),
@@ -16,6 +18,11 @@ AC_DEFUN([ACVT_LIBC],
 	AC_MSG_CHECKING([for LIBC's pathname])
 
 	AS_IF([test x"$libc_pathname" = x],
+	[
+		AS_IF([test "$PLATFORM" = "bgp"],
+		[
+			libc_pathname="/lib/libc.so.6"
+		],
 		[
 			rm -f conftest
 			AC_TRY_LINK([], [],
@@ -37,6 +44,7 @@ AC_DEFUN([ACVT_LIBC],
 				libc_error="yes"
 			])
 		])
+	])
 
 	AS_IF([test x"$libc_pathname" != x -a x"$libc_error" = "xno"],
 	[

ompi/contrib/vt/vt/config/m4/acinclude.libwrap.m4

@@ -10,6 +10,8 @@ AC_DEFUN([ACVT_LIBWRAP],
 	force_libcwrap="no"
 	force_iowrap="no"
 
+	AC_REQUIRE([ACVT_PLATFORM])
+
 	AC_ARG_ENABLE(libtrace,
 		AC_HELP_STRING([--enable-libtrace=LIST],
 			[enable library tracing support (gen,libc,io), default: automatically by configure]),
@@ -41,9 +43,18 @@ AC_DEFUN([ACVT_LIBWRAP],
 	])
 
 	AS_IF([test x"$check_libwrap" != "xno"],
+	[
+		AS_IF([test "$PLATFORM" = "bgp" -a x"$enable_shared" = "xno"],
+		[
+			AC_MSG_NOTICE([error: library tracing requires building of shared libraries on this platform; re-configure with \`--enable-shared'])
+			libwrap_error="yes"
+		])
+
+		AS_IF([test x"$libwrap_error" = "xno"],
 		[
 			ACVT_DL
 			AS_IF([test x"$have_dl" = "xno"], [libwrap_error="yes"])
+		])
 
 		AS_IF([test x"$libwrap_error" = "xno"],
 		[

ompi/contrib/vt/vt/config/m4/acinclude.pform.m4

@@ -79,6 +79,12 @@ AC_DEFUN([ACVT_PLATFORM],
 		BITMODE=$withval
         ])
 
+
+	AS_IF([test "$PLATFORM" = "bgp"],
+	[
+		CPPFLAGS="$CPPFLAGS -I/bgsys/drivers/ppcfloor/arch/include"
+	])
+
 	AC_SUBST(PLATFORM)
 	AC_SUBST(BITMODE)
 ])

ompi/contrib/vt/vt/config/m4/acinclude.timer.m4

@@ -59,7 +59,6 @@ AC_DEFUN([ACVT_TIMER],
 		timer=TIMER_RTS_GET_TIMEBASE
 		;;
 	bgp)
-		CPPFLAGS="$CPPFLAGS -I/bgsys/drivers/ppcfloor/arch/include"
 		AC_DEFINE([TIMER_BGP_GET_TIMEBASE], [1], [Use `_bgp_GetTimeBase' function])
 		timer=TIMER_BGP_GET_TIMEBASE
 		;;

ompi/contrib/vt/vt/configure.in

@@ -147,14 +147,20 @@ AC_CHECK_FUNCS([asprintf snprintf vasprintf vsnprintf strdup])
 
 # Check for libtool
 ACVT_CONF_TITLE([Libtool])
-AS_IF([test "$PLATFORM" = "bgl" -o "$PLATFORM" = "bgp"],
+
-[
+# Override the configure options '--enable-static' and '--enable-shared' to
-	AS_IF([test x"$enable_shared" = "xyes"],
+# enforce building only static libraries on CrayXT and BlueGene/L platforms.
-	[AC_MSG_WARN([VampirTrace does not support shared libraries on IBM BlueGene; build static libraries])])
+AS_IF([test "$PLATFORM" = "crayxt" -o "$PLATFORM" = "bgl"],
-	enable_shared="no"; enable_static="yes"
+[enable_static="yes"; enable_shared="no"])
-])
+
 AC_PROG_LIBTOOL
 
+# If building of shared libraries is desired on BlueGene/P, add '-Wl,-dy'
+# to linker flags on BlueGene/P platforms to lead libtool to believe that
+# dynamic linking is the default behaviour of the linker.
+AS_IF([test "$PLATFORM" = "bgp" -a x"$enable_shared" = "xyes"],
+[export LDFLAGS="$LDFLAGS -Wl,-dy"])
+
 # Check for OTF
 ACVT_CONF_TITLE([Open Trace Format (OTF)])
 ACVT_OTF

ompi/contrib/vt/vt/extlib/otf/ChangeLog

@@ -2,7 +2,10 @@
 	- enforce building position independent code (PIC) for the static
 	  version of the OTF library
 	  (so it's possible to add these objects to 3rd-party shared libraries)
-	- patched libtool to avoid a bug in detection of the PGI 10 C++ compiler
+	- fixed and enhanced calculation of trace file size in otfinfo
+	- patched libtool:
+	  - to avoid a bug in detection of the PGI 10 C++ compiler
+	  - to detect IBM cross compilers on BlueGene/P
 
 1.6.4openmpi
 	- enhanced reading of process group definition records

ompi/contrib/vt/vt/extlib/otf/tools/otfinfo/otfinfo.c

@@ -11,6 +11,9 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <sys/types.h>
 
 #define PROGRESSBARLEN 20
 
@@ -42,10 +45,12 @@ int main(int argc, char **argv)
   uint64_t          size           = 0;    /*size of the event files*/
   uint64_t          minRead        = 0;    /*variable for progress*/
   uint64_t          currRead       = 0;    /*count of current read bytes*/
+  uint32_t          numStreams     = 0;
   char*             parameter      = NULL; /*reference to the parameters*/
   char*             fileLocation   = NULL; /*path to the tracefiles*/
   OTF_FileManager  *manager        = NULL;
   OTF_Reader       *reader         = NULL;
+  OTF_MasterControl* master        = NULL;
   OTF_HandlerArray *handles        = NULL;
   int showProgress                 = 0;
   int infoLevel                    = 1;    /*level for the output of information (local)*/
@@ -167,18 +172,44 @@ int main(int argc, char **argv)
   /*read definition file*/
   checkVal = OTF_Reader_readDefinitions( reader, handles );
   otfinfo_assert( checkVal != OTF_READ_ERROR );
-
   checkVal = OTF_Reader_readMarkers( reader, handles );
   otfinfo_assert( checkVal != OTF_READ_ERROR );
 
   /*getting the size of the event files*/
-  OTF_Reader_setRecordLimit( reader, 0 );
+  master = OTF_Reader_getMasterControl( reader );
-  checkVal = OTF_Reader_readEvents( reader, handles );
+  otfinfo_assert( master );
-  otfinfo_assert(checkVal != OTF_READ_ERROR);
+
-  checkVal = OTF_Reader_eventBytesProgress( reader, &minRead,
+  numStreams = OTF_MasterControl_getCount( master );
-                                            &currRead, &size);
+  otfinfo_assert( numStreams > 0 );
-  otfinfo_assert( checkVal != OTF_READ_ERROR );
+
-  info.traceFileSize = size;
+  info.traceFileSize = 0;
+  for( i = 0; i < (int)numStreams; i++ )
+  {
+    static char filename[1024];
+    static uint32_t filecomp = OTF_FILECOMPRESSION_COMPRESSED;
+    static struct stat filestat;
+
+    /*get event file name of stream*/
+    OTF_getFilename( fileLocation, i+1, OTF_FILETYPE_EVENT | filecomp,
+                     sizeof(filename), filename );
+
+    /*if stat succeeds, compute total file size*/
+    if( stat( filename, &filestat ) == 0 )
+    {
+      info.traceFileSize += (uint64_t)filestat.st_size;
+    }
+    /*otherwise, re-try with uncompressed file*/
+    else
+    {
+      if( filecomp == OTF_FILECOMPRESSION_COMPRESSED )
+      {
+        filecomp = OTF_FILECOMPRESSION_UNCOMPRESSED;
+        i--; continue;
+      }
+    }
+
+    filecomp = OTF_FILECOMPRESSION_COMPRESSED;
+  }
 
   /*printing the results and cleanup*/
   if( 0 < infoLevel )
@@ -339,15 +370,15 @@ static void show_info_level_1( definitionInfoT *info )
   }
   switch( i )
   {
-    case 0: unitTimer = "s";break;
+    case 0: unitTimer = "Hz";break;
-    case 1: unitTimer = "ms";break;
+    case 1: unitTimer = "KHz";break;
-    case 2: unitTimer = "µs";break;
+    case 2: unitTimer = "MHz";break;
-    default: unitTimer = "ns";break;
+    default: unitTimer = "GHz";break;
   }
 
   /*formating the size of the event files*/
   i = 0;
-  while( (fileSize / 1024 >= 1.0) && (i < 4) )
+  while( (fileSize / 1024 >= 1.0) && (i < 5) )
   {
     fileSize /= 1024;
     i++;
@@ -357,6 +388,7 @@ static void show_info_level_1( definitionInfoT *info )
     case 0: unitFileSize = "Bytes"; break;
     case 1: unitFileSize = "KB"; break;
     case 2: unitFileSize = "MB"; break;
+    case 3: unitFileSize = "GB"; break;
     default: unitFileSize = "TB";break;
   }
 

ompi/contrib/vt/vt/extlib/otf/tools/otfprofile/Handler.cpp

@@ -119,7 +119,18 @@ int handleDefProcess(void* firsthandlerarg, uint32_t streamid,
 	}
 	else
 	{
-		gd_ptr->sum_container.adddef_Proc(p_def_key, strdup(name));
+
+		char* dup= strdup( name );
+		char* p= dup;
+		while ( '\0' != *p ) {
+
+			if ( '_' == *p ) *p= ' ';
+			if ( '\\' == *p ) *p= ' ';
+
+			p++;
+		}
+
+		gd_ptr->sum_container.adddef_Proc(p_def_key, dup );
 	}
 	Process p;
 

ompi/contrib/vt/vt/extlib/otf/tools/otfprofile/otfprofile.cpp

@@ -554,7 +554,9 @@ int main( int argc, const char** argv )
 			}
 
 #			ifdef _OPENMP
+			if(status) {
 #				pragma omp barrier
+			}
 #			endif
 			if( (read <= 0) && (ready == false) ) {
 #				ifdef _OPENMP
@@ -585,11 +587,10 @@ int main( int argc, const char** argv )
 					}
 				}
 			}
-#			ifdef _OPENMP
+
-#				pragma omp barrier
-#			endif
 			if(status && ready == false) {
 #				ifdef _OPENMP
+#					pragma omp barrier
 #					pragma omp single nowait
 #				endif
 				{
@@ -599,6 +600,7 @@ int main( int argc, const char** argv )
 				}
 			}
 		}
+
 		OTF_Reader_setRecordLimit( reader, (uint64_t) OTF_READ_MAXRECORDS );
 			
 		OTF_Reader_close(reader);

ompi/contrib/vt/vt/tools/vtunify/mpi/vt_unify_mpi.c

@@ -248,23 +248,30 @@ VT_MPI_INT VTUnify_MPI_Type_free( VTUnify_MPI_Datatype * utype )
 
 VT_MPI_INT VTUnify_MPI_Type_struct( VT_MPI_INT count,
                                     VT_MPI_INT * array_of_blocklengths,
-                                    VTUnify_MPI_Aint * array_of_displacements,
+                                    VTUnify_MPI_Aint * array_of_udisplacements,
                                     VTUnify_MPI_Datatype * array_of_utypes,
                                     VTUnify_MPI_Datatype * newutype )
 {
    VT_MPI_INT error, i;
+   MPI_Aint * array_of_displacements;
    MPI_Datatype * array_of_types;
    MPI_Datatype newtype;
 
+   array_of_displacements = (MPI_Aint*)malloc( count * sizeof( MPI_Aint ));
+   assert( array_of_displacements );
+   for( i = 0; i < count; i++ )
+      array_of_displacements[i] = (MPI_Aint)array_of_udisplacements[i];
+
    array_of_types = (MPI_Datatype*)malloc( count * sizeof( MPI_Datatype ));
    assert( array_of_types );
    for( i = 0; i < count; i++ )
       array_of_types[i] = get_mpi_type( array_of_utypes[i] );
 
    error = MPI_Type_struct( count, array_of_blocklengths,
-                            (MPI_Aint*)array_of_displacements, array_of_types,
+                            array_of_displacements, array_of_types,
                             &newtype );
 
+   free( array_of_displacements );
    free( array_of_types );
 
    if( error == MPI_SUCCESS )

ompi/contrib/vt/vt/tools/vtunify/mpi/vt_unify_mpi.h

@@ -95,7 +95,7 @@ EXTERN VT_MPI_INT VTUnify_MPI_Type_free( VTUnify_MPI_Datatype * utype );
 
 EXTERN VT_MPI_INT VTUnify_MPI_Type_struct( VT_MPI_INT count,
                                            VT_MPI_INT * array_of_blocklengths,
-                                           VTUnify_MPI_Aint * array_of_displacements,
+                                           VTUnify_MPI_Aint * array_of_udisplacements,
                                            VTUnify_MPI_Datatype * array_of_utypes,
                                            VTUnify_MPI_Datatype * newutype );
 

ompi/contrib/vt/vt/tools/vtwrapper/vt_wrapper.cc

@@ -981,7 +981,8 @@ Wrapper::run()
                std::cout << "+++ rename " << m_pConfig->m_vecOpari_ModObjFiles[i]
                          << " to " << target << std::endl;
 
-            if( rename( m_pConfig->m_vecOpari_ModObjFiles[i].c_str(),
+            if( m_pConfig->m_bCompOnly &&
+                rename( m_pConfig->m_vecOpari_ModObjFiles[i].c_str(),
                         target.c_str() ) == -1 )
             {
                std::cerr << ExeName << ": could not rename "

ompi/contrib/vt/vt/vtlib/vt_otf_gen.c

@@ -133,7 +133,7 @@ struct VTGen_struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 } VTBuf_Entry_Base;
 
 /* BUF_ENTRY_TYPE__DefinitionComment */
@@ -141,9 +141,9 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
-  char* comment;
+  char comment[1];
 } VTBuf_Entry_DefinitionComment;
 
 /* BUF_ENTRY_TYPE__DefSclFile */
@@ -151,10 +151,10 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t fid;
-  char*   fname;
+  char     fname[1];
 } VTBuf_Entry_DefSclFile;
 
 /* BUF_ENTRY_TYPE__DefScl */
@@ -162,7 +162,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t sid;
   uint32_t fid;
@@ -174,10 +174,10 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t gid;
-  char*    gname;
+  char     gname[1];
 } VTBuf_Entry_DefFileGroup;
 
 /* BUF_ENTRY_TYPE__DefFile */
@@ -185,11 +185,11 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t fid;
-  char*    fname;
   uint32_t gid;
+  char     fname[1];
 } VTBuf_Entry_DefFile;
 
 /* BUF_ENTRY_TYPE__DefFunctionGroup */
@@ -197,10 +197,10 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t rdid;
-  char*   rdesc;    
+  char     rdesc[1];
 } VTBuf_Entry_DefFunctionGroup;
 
 /* BUF_ENTRY_TYPE__DefFunction */
@@ -208,12 +208,12 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t rid;
-  char*   rname;
   uint32_t rdid;
   uint32_t sid;
+  char     rname[1];
 } VTBuf_Entry_DefFunction;
 
 /* BUF_ENTRY_TYPE__DefCollectiveOperation */
@@ -221,11 +221,11 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t cid;
-  char*   cname;
   uint32_t ctype;
+  char     cname[1];
 } VTBuf_Entry_DefCollectiveOperation;
 
 /* BUF_ENTRY_TYPE__DefCounterGroup */
@@ -233,10 +233,10 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t gid;
-  char*   gname;
+  char     gname[1];
 } VTBuf_Entry_DefCounterGroup;
 
 /* BUF_ENTRY_TYPE__DefCounter */
@@ -244,13 +244,13 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t cid;
-  char*   cname;
   uint32_t cprop;
   uint32_t gid;
-  char*   cunit;
+  char     cunit[100];
+  char     cname[1];
 } VTBuf_Entry_DefCounter;
 
 /* BUF_ENTRY_TYPE__DefProcessGroup */
@@ -258,12 +258,12 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint32_t  cid;
-  char*    grpn;
+  char      grpn[100];
   uint32_t  grpc;
-  uint32_t* grpv;
+  uint32_t  grpv[1];
 } VTBuf_Entry_DefProcessGroup;
 
 /* BUF_ENTRY_TYPE__DefMarker */
@@ -271,11 +271,11 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-   uint8_t length;
+  uint32_t length;
 
   uint32_t mid;
-  char* mname;
   uint32_t mtype;
+  char     mname[1];
 } VTBuf_Entry_DefMarker;
 
 /* BUF_ENTRY_TYPE__Enter / BUF_ENTRY_TYPE__Leave */
@@ -283,7 +283,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t rid;
@@ -297,7 +297,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint64_t etime;
@@ -311,7 +311,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint64_t hid;
@@ -321,7 +321,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t fid;
@@ -336,7 +336,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t cid;
@@ -348,10 +348,10 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
-  char* comment;
+  char     comment[1];
 } VTBuf_Entry_Comment;
 
 /* BUF_ENTRY_TYPE__Marker */
@@ -359,11 +359,11 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t mid;
-  char* mtext;
+  char     mtext[1];
 } VTBuf_Entry_Marker;
 
 /* BUF_ENTRY_TYPE__SendMsg / BUF_ENTRY_TYPE__RecvMsg */
@@ -371,7 +371,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t pid;
@@ -386,7 +386,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint64_t etime;
@@ -403,7 +403,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time; 
   uint32_t opid; 
@@ -419,7 +419,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t rpid;
@@ -433,7 +433,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t rid;
@@ -447,7 +447,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t peer;
@@ -464,7 +464,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t cid;
@@ -480,7 +480,7 @@ typedef struct
 typedef struct
 {
   VTBuf_EntryTypes type;
-  uint8_t length;
+  uint32_t length;
 
   uint64_t time;
   uint32_t fid;
@@ -583,7 +583,10 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
   buffer_t p;
   int i;
 
+  /* buffer empty? */
   if(gen->buf->pos == gen->buf->mem) return;
+
+  /* intermediate flush and max. buffer flushes reached? */
   if(!lastFlush && gen->flushcntr == 0) return;
 
   /* Disable I/O tracing */
@@ -706,7 +709,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	 OTF_WStream_writeDefinitionComment(gen->filestream,
 	   entry->comment);
 
-	 free(entry->comment);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefSclFile:
@@ -718,7 +720,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->fid,
 	   entry->fname);
 
-	 free(entry->fname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefScl:
@@ -742,7 +743,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->gid,
 	   entry->gname);
 
-	 free(entry->gname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefFile:
@@ -755,7 +755,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->fname,
 	   entry->gid);
 
-	 free(entry->fname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefFunctionGroup:
@@ -767,7 +766,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->rdid,
 	   entry->rdesc);
 
-	 free(entry->rdesc);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefFunction:
@@ -781,7 +779,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->rdid,
 	   entry->sid);
 
-	 free(entry->rname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefCollectiveOperation:
@@ -794,7 +791,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->cname,
 	   entry->ctype);
 
-	 free(entry->cname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefCounterGroup:
@@ -806,7 +802,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->gid,
 	   entry->gname);
 
-	 free(entry->gname);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefCounter:
@@ -821,8 +816,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->gid,
 	   entry->cunit);
 
-	 free(entry->cname);
-	 free(entry->cunit);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefProcessGroup:
@@ -836,8 +829,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->grpc,
 	   entry->grpv);
 
-	 free(entry->grpn);
-	 free(entry->grpv);
 	 break;
        }
        case BUF_ENTRY_TYPE__DefMarker:
@@ -850,7 +841,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   entry->mname,
 	   entry->mtype);
 
-	 free(entry->mname);
 	 break;
        }
        case BUF_ENTRY_TYPE__Enter:
@@ -965,7 +955,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   65536 * gen->tid + vt_my_trace + 1,
 	   entry->comment);
 
-	 free(entry->comment);
 	 break;
        }
        case BUF_ENTRY_TYPE__Marker:
@@ -978,7 +967,6 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
 	   65536 * gen->tid + vt_my_trace + 1,
 	   entry->mid, entry->mtext);
 
-	 free(entry->mtext);
 	 break;
        }
        case BUF_ENTRY_TYPE__SendMsg:
@@ -1167,8 +1155,9 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
      if(!end_flush_marked &&
 	p + ((VTBuf_Entry_Base*)p)->length >= gen->buf->pos)
      {
-       /* mark end of flush */
+       /* mark end of flush, if it's not the last (invisible) flush and 
-       if(!lastFlush)
+          max flushes not reached */
+       if(!lastFlush && gen->flushcntr > 1)
        {
 	 uint64_t flush_etime = vt_pform_wtime();
 	 vt_exit_flush(&flush_etime);
@@ -1196,7 +1185,7 @@ void VTGen_flush(VTGen* gen, uint8_t lastFlush,
     int max_flushes = vt_env_max_flushes();
     vt_cntl_msg(1, "Maximum number of buffer flushes reached "
 		"(VT_MAX_FLUSHES=%d)", max_flushes);
-    vt_trace_off(0, 1);
+    vt_trace_off(1, 1);
     vt_def_comment("__VT_COMMENT__ WARNING: This trace is "
 		   "incomplete, because the maximum number of "
 		   "buffer flushes was reached. "
@@ -1413,6 +1402,8 @@ void VTGen_destroy(VTGen* gen)
 
 uint8_t VTGen_get_buflevel(VTGen* gen)
 {
+  VTGEN_CHECK(gen);
+
   return (uint8_t)(((gen->buf->pos - gen->buf->mem) * 100) / gen->buf->size);
 }
 
@@ -1422,13 +1413,13 @@ uint8_t VTGen_get_buflevel(VTGen* gen)
 
 /* - Definition records - */
 
-void VTGen_write_DEFINITION_COMMENT(VTGen* gen,
+void VTGen_write_DEFINITION_COMMENT(VTGen* gen, const char* comment)
-				     const char* comment)
 {
   VTBuf_Entry_DefinitionComment* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefinitionComment));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefinitionComment) +
+                        (strlen(comment) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1438,7 +1429,7 @@ void VTGen_write_DEFINITION_COMMENT(VTGen* gen,
 
   new_entry->type    = BUF_ENTRY_TYPE__DefinitionComment;
   new_entry->length  = length;
-  new_entry->comment = strdup(comment);
+  strcpy(new_entry->comment, comment);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1449,8 +1440,9 @@ void VTGen_write_DEF_SCL_FILE(VTGen* gen,
 {
   VTBuf_Entry_DefSclFile* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefSclFile));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefSclFile) +
+                        (strlen(fname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1461,7 +1453,7 @@ void VTGen_write_DEF_SCL_FILE(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefSclFile;
   new_entry->length = length;
   new_entry->fid    = fid;
-  new_entry->fname  = strdup(fname);
+  strcpy(new_entry->fname, fname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1473,7 +1465,7 @@ void VTGen_write_DEF_SCL(VTGen* gen,
 {
   VTBuf_Entry_DefScl* new_entry;
 
-  uint8_t length =
+  uint32_t length =
     VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefScl));
 
   VTGEN_CHECK(gen);
@@ -1496,8 +1488,9 @@ void VTGen_write_DEF_FILE_GROUP(VTGen* gen,
 {
   VTBuf_Entry_DefFileGroup* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefFileGroup));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefFileGroup) +
+                        (strlen(gname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1508,7 +1501,7 @@ void VTGen_write_DEF_FILE_GROUP(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefFileGroup;
   new_entry->length = length;
   new_entry->gid    = gid;
-  new_entry->gname  = strdup(gname);
+  strcpy(new_entry->gname, gname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1520,8 +1513,9 @@ void VTGen_write_DEF_FILE(VTGen* gen,
 {
   VTBuf_Entry_DefFile* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefFile));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefFile) +
+                        (strlen(fname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1532,8 +1526,8 @@ void VTGen_write_DEF_FILE(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefFile;
   new_entry->length = length;
   new_entry->fid    = fid;
-  new_entry->fname  = strdup(fname);
   new_entry->gid    = gid;
+  strcpy(new_entry->fname, fname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1544,8 +1538,9 @@ void VTGen_write_DEF_FUNCTION_GROUP(VTGen* gen,
 {
   VTBuf_Entry_DefFunctionGroup* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefFunctionGroup));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefFunctionGroup) +
+                        (strlen(rdesc) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1556,7 +1551,7 @@ void VTGen_write_DEF_FUNCTION_GROUP(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefFunctionGroup;
   new_entry->length = length;
   new_entry->rdid   = rdid;
-  new_entry->rdesc  = strdup(rdesc);
+  strcpy(new_entry->rdesc, rdesc);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1569,8 +1564,9 @@ void VTGen_write_DEF_FUNCTION(VTGen* gen,
 {
   VTBuf_Entry_DefFunction* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefFunction));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefFunction) +
+                        (strlen(rname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1581,9 +1577,9 @@ void VTGen_write_DEF_FUNCTION(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefFunction;
   new_entry->length = length;
   new_entry->rid    = rid;
-  new_entry->rname  = strdup(rname);
   new_entry->rdid   = rdid;
   new_entry->sid    = sid;
+  strcpy(new_entry->rname, rname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1595,8 +1591,9 @@ void VTGen_write_DEF_COLLECTIVE_OPERATION(VTGen* gen,
 {
   VTBuf_Entry_DefCollectiveOperation* new_entry;
 
-  uint8_t length = 
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefCollectiveOperation));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefCollectiveOperation) +
+                        (strlen(cname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1607,8 +1604,8 @@ void VTGen_write_DEF_COLLECTIVE_OPERATION(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefCollectiveOperation;
   new_entry->length = length;
   new_entry->cid    = cid;
-  new_entry->cname  = strdup(cname);
   new_entry->ctype  = ctype;
+  strcpy(new_entry->cname, cname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1619,8 +1616,9 @@ void VTGen_write_DEF_COUNTER_GROUP(VTGen* gen,
 {
   VTBuf_Entry_DefCounterGroup* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefCounterGroup));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefCounterGroup) +
+                        (strlen(gname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1631,7 +1629,7 @@ void VTGen_write_DEF_COUNTER_GROUP(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefCounterGroup;
   new_entry->length = length;
   new_entry->gid    = gid;
-  new_entry->gname  = strdup(gname);
+  strcpy(new_entry->gname, gname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1645,8 +1643,9 @@ void VTGen_write_DEF_COUNTER(VTGen* gen,
 {
   VTBuf_Entry_DefCounter* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefCounter));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefCounter) +
+                        (strlen(cname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1657,10 +1656,11 @@ void VTGen_write_DEF_COUNTER(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefCounter;
   new_entry->length = length;
   new_entry->cid    = cid;
-  new_entry->cname  = strdup(cname);
   new_entry->cprop  = cprop;
   new_entry->gid    = gid;
-  new_entry->cunit  = strdup(cunit);
+  strncpy(new_entry->cunit, cunit, sizeof(new_entry->cunit)-1);
+  new_entry->cunit[sizeof(new_entry->cunit)-1] = '\0';
+  strcpy(new_entry->cname, cname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1673,8 +1673,9 @@ void VTGen_write_DEF_PROCESS_GROUP(VTGen* gen,
 {
   VTBuf_Entry_DefProcessGroup* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefProcessGroup));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefProcessGroup) +
+                        (grpc > 0 ? (grpc - 1) * sizeof(uint32_t) : 0 )));
 
   VTGEN_CHECK(gen);
 
@@ -1685,12 +1686,10 @@ void VTGen_write_DEF_PROCESS_GROUP(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefProcessGroup;
   new_entry->length = length;
   new_entry->cid    = cid;
-  new_entry->grpn   = strdup(grpn);
+  strncpy(new_entry->grpn, grpn, sizeof(new_entry->grpn)-1);
+  new_entry->grpn[sizeof(new_entry->grpn)-1] = '\0';
   new_entry->grpc   = grpc;
-  new_entry->grpv   = (uint32_t*)calloc(grpc, sizeof(uint32_t));
+  if( grpc > 0 )
-  if(new_entry->grpv == NULL)
-    vt_error();
-
     memcpy(new_entry->grpv, grpv, grpc * sizeof(uint32_t));
 
   VTGEN_JUMP(gen, length);
@@ -1705,8 +1704,9 @@ void VTGen_write_DEF_MARKER(VTGen* gen,
 {
   VTBuf_Entry_DefMarker* new_entry;
 
-  uint8_t length =
+  uint32_t length =
-    VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_DefMarker));
+    VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_DefMarker) +
+                        (strlen(mname) * sizeof(char))));
 
   VTGEN_CHECK(gen);
 
@@ -1717,8 +1717,8 @@ void VTGen_write_DEF_MARKER(VTGen* gen,
   new_entry->type   = BUF_ENTRY_TYPE__DefMarker;
   new_entry->length = length;
   new_entry->mid    = mid;
-  new_entry->mname  = vt_strdup(mname);
   new_entry->mtype  = mtype;
+  strcpy(new_entry->mname, mname);
 
   VTGEN_JUMP(gen, length);
 }
@@ -1738,9 +1738,8 @@ void VTGen_write_ENTER(VTGen* gen, uint64_t* time, uint32_t rid, uint32_t sid,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) +
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) +
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0 )));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -1772,9 +1771,8 @@ void VTGen_write_LEAVE(VTGen* gen, uint64_t* time, uint32_t rid, uint32_t sid,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) +
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0)));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -1809,9 +1807,8 @@ void VTGen_write_FILE_OPERATION(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_FileOperation* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-    
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FileOperation));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FileOperation));
 
     *etime -= *time;
     VTGEN_ALLOC_EVENT(gen, length);
@@ -1874,9 +1871,8 @@ void VTGen_write_BEGIN_FILE_OPERATION(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_BeginFileOperation* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_BeginFileOperation));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_BeginFileOperation));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -1902,9 +1898,8 @@ void VTGen_write_END_FILE_OPERATION(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_EndFileOperation* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_EndFileOperation));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_EndFileOperation));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -1969,9 +1964,8 @@ void VTGen_write_COUNTER(VTGen* gen, uint64_t* time, uint32_t cid,
   {
     VTBuf_Entry_Counter* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_Counter));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_Counter));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -1998,7 +1992,7 @@ void VTGen_write_COMMENT(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_Comment* new_entry;
 
-    uint8_t length =
+    uint32_t length =
       VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_Comment));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -2008,7 +2002,7 @@ void VTGen_write_COMMENT(VTGen* gen, uint64_t* time,
     new_entry->type    = BUF_ENTRY_TYPE__Comment;
     new_entry->length  = length;
     new_entry->time    = *time;
-    new_entry->comment = strdup(comment);
+    strcpy(new_entry->comment, comment);
 
     VTGEN_JUMP(gen, length);
   }
@@ -2025,7 +2019,7 @@ void VTGen_write_MARKER(VTGen* gen, uint64_t* time, uint32_t mid,
   {
     VTBuf_Entry_Marker* new_entry;
 
-    uint8_t length =
+    uint32_t length =
       VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_Marker));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -2036,7 +2030,7 @@ void VTGen_write_MARKER(VTGen* gen, uint64_t* time, uint32_t mid,
     new_entry->length  = length;
     new_entry->time    = *time;
     new_entry->mid     = mid;
-    new_entry->mtext   = vt_strdup(mtext);
+    strcpy(new_entry->mtext, mtext);
 
     VTGEN_JUMP(gen, length);
   }
@@ -2053,9 +2047,8 @@ void VTGen_write_SEND_MSG(VTGen* gen, uint64_t* time, uint32_t pid,
   {
     VTBuf_Entry_SendRecvMsg* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_SendRecvMsg));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_SendRecvMsg));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2086,9 +2079,8 @@ void VTGen_write_RECV_MSG(VTGen* gen, uint64_t* time, uint32_t pid,
   {
     VTBuf_Entry_SendRecvMsg* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-    
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_SendRecvMsg));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_SendRecvMsg));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2120,9 +2112,8 @@ void VTGen_write_COLLECTIVE_OPERATION(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_CollectiveOperation* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-    
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_CollectiveOperation));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_CollectiveOperation));
 
     *etime -= *time;
     VTGEN_ALLOC_EVENT(gen, length);
@@ -2159,9 +2150,8 @@ void VTGen_write_RMA_PUT(VTGen* gen, uint64_t* time, uint32_t opid,
   {
     VTBuf_Entry_RMAPutGet* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2195,9 +2185,8 @@ void VTGen_write_RMA_PUTRE(VTGen* gen, uint64_t* time, uint32_t opid,
   {
     VTBuf_Entry_RMAPutGet* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2226,9 +2215,8 @@ void VTGen_write_RMA_GET(VTGen* gen, uint64_t* time, uint32_t opid,
   {
     VTBuf_Entry_RMAPutGet* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAPutGet));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2257,9 +2245,8 @@ void VTGen_write_RMA_END(VTGen* gen, uint64_t* time, uint32_t rpid,
   {
     VTBuf_Entry_RMAEnd* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAEnd));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_RMAEnd));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2288,9 +2275,8 @@ void VTGen_write_ENTER_STAT(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0)));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -2319,9 +2305,8 @@ void VTGen_write_EXIT_STAT(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0)));
 
     VTGEN_ALLOC_EVENT(gen, length);
@@ -2350,9 +2335,8 @@ void VTGen_write_ENTER_FLUSH(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0)));
 
     /* NB: No VTGEN_ALLOC_EVENT since space reserved at buffer creation */
@@ -2381,9 +2365,8 @@ void VTGen_write_EXIT_FLUSH(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_EnterLeave* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
-    length = VTGEN_ALIGN_LENGTH((sizeof(VTBuf_Entry_EnterLeave) + 
                           (metc > 0 ? (metc - 1) * sizeof(uint64_t) : 0)));
 
     /* NB: No VTGEN_ALLOC_EVENT since space reserved at buffer creation */
@@ -2416,9 +2399,8 @@ void VTGen_write_FUNCTION_SUMMARY(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_FunctionSummary* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FunctionSummary));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FunctionSummary));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2446,9 +2428,8 @@ void VTGen_write_MESSAGE_SUMMARY(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_MessageSummary* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_MessageSummary));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_MessageSummary));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2479,9 +2460,8 @@ void VTGen_write_COLLECTIVE_OPERATION_SUMMARY(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_CollectiveOperationSummary* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_CollectiveOperationSummary));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_CollectiveOperationSummary));
 
     VTGEN_ALLOC_EVENT(gen, length);
 
@@ -2511,9 +2491,8 @@ void VTGen_write_FILE_OPERATION_SUMMARY(VTGen* gen, uint64_t* time,
   {
     VTBuf_Entry_FileOperationSummary* new_entry;
 
-    uint8_t length;
+    uint32_t length =
-
+      VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FileOperationSummary));
-    length = VTGEN_ALIGN_LENGTH(sizeof(VTBuf_Entry_FileOperationSummary));
 
     VTGEN_ALLOC_EVENT(gen, length);
 

ompi/contrib/vt/vt/vtlib/vt_otf_trc.c

@@ -941,31 +941,22 @@ void vt_trace_off(uint8_t mark, uint8_t permanent)
 
   if ( vt_is_alive &&
        VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) != VT_TRACE_OFF_PERMANENT )
-  {
-    uint64_t time;
-
-    if ( permanent )
-    {
-      /* shutdown call stack */
-      while(VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD]) > 0) {
-	time = vt_pform_wtime();
-	vt_exit(&time);
-      }
-
-      VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) = VT_TRACE_OFF_PERMANENT;
-
-      vt_cntl_msg(1, "Tracing switched off permanently");
-    }
-    else
-    {
-      if ( VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) == VT_TRACE_ON )
   {
     if ( mark )
     {
+      uint64_t time;
       time = vt_pform_wtime();
       vt_enter(&time, vt_trc_regid[VT__TRC_OFF]);
     }
 
+    if ( permanent )
+    {
+      VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) = VT_TRACE_OFF_PERMANENT;
+
+      vt_cntl_msg(1, "Tracing switched off permanently");
+    }
+    else if ( VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) == VT_TRACE_ON )
+    {
       /* store current call stack level */
       VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) = VT_TRACE_OFF;
       VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]) =
@@ -975,7 +966,6 @@ void vt_trace_off(uint8_t mark, uint8_t permanent)
       VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]));
     }
   }
-  }
 }
 
 uint8_t vt_is_trace_on()
@@ -1834,8 +1824,10 @@ uint8_t vt_enter(uint64_t* time, uint32_t rid) {
   if (VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) ==
       VT_TRACE_OFF_PERMANENT) return 0;
 
+  VTTHRD_STACK_PUSH(VTThrdv[VT_MY_THREAD]);
+
   do_trace = ((VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) == VT_TRACE_ON) &&
-	      (VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD]) < max_stack_depth));
+	      (VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD]) <= max_stack_depth));
 
 #if !defined(VT_DISABLE_RFG)
   if( !RFG_Regions_stackPush(VTTHRD_RFGREGIONS(VTThrdv[VT_MY_THREAD]),
@@ -1924,11 +1916,6 @@ uint8_t vt_enter(uint64_t* time, uint32_t rid) {
     UPDATE_RUSAGE(time);
   }
 
-  /* push call stack, if tracing was not disabled by this event
-     (max buffer flushes reached) */
-  if (VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) != VT_TRACE_OFF_PERMANENT)
-    VTTHRD_STACK_PUSH(VTThrdv[VT_MY_THREAD]);
-
   return do_trace;
 }
 
@@ -1944,8 +1931,10 @@ void vt_exit(uint64_t* time) {
   if (VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) ==
       VT_TRACE_OFF_PERMANENT) return;
 
+  VTTHRD_STACK_POP(VTThrdv[VT_MY_THREAD]);
+
   do_trace = ((VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) == VT_TRACE_ON) &&
-	      (VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD])+1 < max_stack_depth));
+	      (VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD])+1 <= max_stack_depth));
 
 #if !defined(VT_DISABLE_RFG)
   if (!RFG_Regions_stackPop(VTTHRD_RFGREGIONS(VTThrdv[VT_MY_THREAD]),
@@ -1995,11 +1984,6 @@ void vt_exit(uint64_t* time) {
 			0, NULL);
 #   endif /* VT_METR */
   }
-
-  /* pop call stack, if tracing was not disabled by this event
-     (max buffer flushes reached) */
-  if (VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) != VT_TRACE_OFF_PERMANENT)
-    VTTHRD_STACK_POP(VTThrdv[VT_MY_THREAD]);
 }
 
 /* -- File I/O -- */