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

@@ -5,6 +5,8 @@ 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]),
@@ -12,9 +14,18 @@ AC_DEFUN([ACVT_JAVA],
 
 	AS_IF([test x"$check_java" = "xyes"],
 	[
-		AC_MSG_CHECKING([whether we can build shared libraries])
+		AS_IF([test "$PLATFORM" = "bgp"],
-		AS_IF([test x"$enable_shared" = "xyes"],
+		[
-		[AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no]); java_error="yes"])
+			AC_MSG_NOTICE([error: Java tracing not supported on this platform])
+			java_error="yes"
+		])
+
+		AS_IF([test x"$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]),
@@ -17,24 +19,30 @@ AC_DEFUN([ACVT_LIBC],
 
 	AS_IF([test x"$libc_pathname" = x],
 	[
-		rm -f conftest
+		AS_IF([test "$PLATFORM" = "bgp"],
-		AC_TRY_LINK([], [],
 		[
-			AS_IF([test -r "conftest"],
+			libc_pathname="/lib/libc.so.6"
+		],
+		[
+			rm -f conftest
+			AC_TRY_LINK([], [],
 			[
-				libc_pathname=`ldd conftest 2>/dev/null | grep "libc\." | \
+				AS_IF([test -r "conftest"],
-				               sed -e "s/.*=>//"                          \
+				[
-				                   -e "s/[[\(].*[\)]]//"                  \
+					libc_pathname=`ldd conftest 2>/dev/null | grep "libc\." | \
-				                   -e "s/[[	 ]]//g"                   \
+					               sed -e "s/.*=>//"                          \
-				                   -e "s%^[[^/]].*%%"                   | \
+					                   -e "s/[[\(].*[\)]]//"                  \
-				               head -n1`
+					                   -e "s/[[	 ]]//g"                   \
+					                   -e "s%^[[^/]].*%%"                   | \
+					               head -n1`
+				],
+				[
+					libc_error="yes"
+				])
 			],
 			[
 				libc_error="yes"
 			])
-		],
-		[
-			libc_error="yes"
 		])
 	])
 

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]),
@@ -42,8 +44,17 @@ AC_DEFUN([ACVT_LIBWRAP],
 
 	AS_IF([test x"$check_libwrap" != "xno"],
 	[
-		ACVT_DL
+		AS_IF([test "$PLATFORM" = "bgp" -a x"$enable_shared" = "xno"],
-		AS_IF([test x"$have_dl" = "xno"], [libwrap_error="yes"])
+		[
+			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_trc.c

@@ -942,38 +942,28 @@ 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 ( mark )
+    {
+      uint64_t time;
+      time = vt_pform_wtime();
+      vt_enter(&time, vt_trc_regid[VT__TRC_OFF]);
+    }
 
     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
+    else if ( VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) == VT_TRACE_ON )
     {
-      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;
-	if ( mark )
+      VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]) =
-	{
+      VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD]);
-	  time = vt_pform_wtime();
-	  vt_enter(&time, vt_trc_regid[VT__TRC_OFF]);
-	}
 
-	/* store current call stack level */
+      vt_cntl_msg(2, "Tracing switched off at call stack level (%i)",
-	VTTHRD_TRACE_STATUS(VTThrdv[VT_MY_THREAD]) = VT_TRACE_OFF;
+      VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]));
-	VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]) =
-	  VTTHRD_STACK_LEVEL(VTThrdv[VT_MY_THREAD]);
-
-	vt_cntl_msg(2, "Tracing switched off at call stack level (%i)",
-		    VTTHRD_STACK_LEVEL_AT_OFF(VTThrdv[VT_MY_THREAD]));
-      }
     }
   }
 }
@@ -1465,7 +1455,7 @@ static void vt_write_def_header()
 
 #if defined(VT_THRD_PTHREAD)
   /* VT_PTHREAD_REUSE */
-  vt_def_comment("__VT_COMMENT__ VT_PTHREAD_REUSE: %s",
+  vt_def_comment("__VT_COMMENT__  VT_PTHREAD_REUSE: %s",
                  vt_env_pthread_reuse() ? "yes" : "no");
 #endif /* VT_THRD_PTHREAD */
 
@@ -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 -- */