o Added option to print memory usage on entry to profiling functions.

  To enable, set in GPTL build:
  - HAVE_BACKTRACE=yes in macros.make
  Then in application:
  - ret = GPTLsetoption (GPTLdopr_memusage, 1)
  - compile with -finstrument-functions -g -rdynamic
o Removed obsolete macros.make.sicortex

configure.ac

@@ -24,6 +24,9 @@ dnl AC_FUNC_VPRINTF
 AC_CHECK_LIB(papi,PAPI_library_init,[echo "PAPI library found: OK to set HAVE_PAPI=yes"],
                                     [echo "PAPI library not found: HAVE_PAPI=no"])
 
+AC_CHECK_FUNC(backtrace_symbols,[echo "backtrace_symbols found: OK to set HAVE_BACKTRACE=yes"],
+				[echo "backtrace_symbols NOT found: HAVE_BACKTRACE=no"])
+
 unset usempich;
 unset usempi;
 AC_CHECK_LIB(mpich,MPI_Init,[echo "libmpich.a found: OK to set HAVE_MPI=yes";usempich=yes],

ftests/Makefile

@@ -103,6 +103,10 @@ overhead: overhead.F90 ../lib$(LIBNAME).a
 toomanythreads: toomanythreads.F90
 	$(FC) $(FFLAGS) -o $@ $< $(LDFLAGS)
 
+# Hidden executable testbacktrace requires some strange flags so don't build it unless asked
+testbacktrace: testbacktrace.F90
+	$(FC) $(FFLAGS) $(INSTRFLAG) -g -rdynamic -o $@ $< $(LDFLAGS)
+
 # Built-in tests to ensure that GPTL is behaving as expected
 # Invoke as "make test" from one dir above
 

gptl.c

@@ -32,6 +32,10 @@
 #include <sys/systemcfg.h>
 #endif
 
+#ifdef HAVE_BACKTRACE
+#include <execinfo.h>
+#endif
+
 #include "private.h"
 #include "gptl.h"
 
@@ -66,6 +70,7 @@ static bool dopr_preamble = true;      /* whether to print preamble info */
 static bool dopr_threadsort = true;    /* whether to print sorted thread stats */
 static bool dopr_multparent = true;    /* whether to print multiple parent info */
 static bool dopr_collision = true;     /* whether to print hash collision info */
+static bool dopr_memusage = false;     /* whether to include memusage print when auto-profiling */
 
 static time_t ref_gettimeofday = -1;   /* ref start point for gettimeofday */
 static time_t ref_clock_gettime = -1;  /* ref start point for clock_gettime */
@@ -224,6 +229,8 @@ static char *clock_source = "UNKNOWN";            /* where clock found */
 static int tablesize = DEFAULT_TABLE_SIZE;  /* per-thread size of hash table (settable parameter) */
 static char *outdir = 0;                    /* dir to write output files to (currently unused) */
 
+#define MSGSIZ 64                           /* max size of msg printed when dopr_memusage=true */
+
 /* VERBOSE is a debugging ifdef local to the rest of this file */
 #undef VERBOSE
 
@@ -310,6 +317,11 @@ int GPTLsetoption (const int option,  /* option */
     if (verbose)
       printf ("%s: boolean dopr_collision = %d\n", thisfunc, val);
     return 0;
+  case GPTLdopr_memusage: 
+    dopr_memusage = (bool) val; 
+    if (verbose)
+      printf ("%s: boolean dopr_memusage = %d\n", thisfunc, val);
+    return 0;
   case GPTLprint_method:
     method = (Method) val; 
     if (verbose)
@@ -2994,6 +3006,12 @@ void __func_trace_enter (const char *function_name,
                          int line_number,
                          void **const user_data)
 {
+  char msg[MSGSIZ];
+
+  if (dopr_memusage) {
+    snprintf (msg, MSGSIZ, "begin %s", function_name);
+    (void) GPTLprint_memusage (msg);
+  }
   (void) GPTLstart (function_name);
 }
 
@@ -3003,20 +3021,61 @@ void __func_trace_exit (const char *function_name,
                         void **const user_data)
 {
   (void) GPTLstop (function_name);
+  if (dopr_memusage) {
+    snprintf (msg, MSGSIZ, "end %s", function_name);
+    (void) GPTLprint_memusage (msg);
+  }
 }
 
 #else
 
 void __cyg_profile_func_enter (void *this_fn,
                                void *call_site)
 {
+#ifdef HAVE_BACKTRACE
+  void *buffer[2];
+  int nptrs;
+  char **strings;
+#endif
+  char msg[MSGSIZ];
+
+  if (dopr_memusage) {
+#ifdef HAVE_BACKTRACE
+    nptrs = backtrace (buffer, 2);
+    strings = backtrace_symbols (buffer, nptrs);
+    snprintf (msg, MSGSIZ, "begin %s", strings[1]);
+    free (strings);
+#else
+    snprintf (msg, MSGSIZ, "begin %lx", (unsigned long) this_fn);
+#endif
+    (void) GPTLprint_memusage (msg);
+  }
   (void) GPTLstart_instr (this_fn);
 }
 
 void __cyg_profile_func_exit (void *this_fn,
                               void *call_site)
 {
+#ifdef HAVE_BACKTRACE
+  void *buffer[2];
+  int nptrs;
+  char **strings;
+#endif
+  char msg[MSGSIZ];
+
   (void) GPTLstop_instr (this_fn);
+
+  if (dopr_memusage) {
+#ifdef HAVE_BACKTRACE
+    nptrs = backtrace (buffer, 2);
+    strings = backtrace_symbols (buffer, nptrs);
+    snprintf (msg, MSGSIZ, "end %s", (char *) strings[1]);
+    free (strings);
+#else
+    snprintf (msg, MSGSIZ, "end %lx", (unsigned long) this_fn);
+#endif
+    (void) GPTLprint_memusage (msg);
+  }
 }
 #endif
 

gptl.h

@@ -33,6 +33,7 @@ typedef enum {
   GPTLdopr_threadsort = 13, /* Print sorted thread stats (true) */
   GPTLdopr_multparent = 14, /* Print multiple parent info (true) */
   GPTLdopr_collision  = 15, /* Print hastable collision info (true) */
+  GPTLdopr_memusage   = 27, /* Call GPTLprint_memusage when auto-instrumented */
   GPTLprint_method    = 16, /* Tree print method: first parent, last parent
 			       most frequent, or full tree (most frequent) */
   GPTLtablesize       = 50, /* per-thread size of hash table */

gptl.h.template

@@ -33,6 +33,7 @@ typedef enum {
   GPTLdopr_threadsort = #GPTLdopr_threadsort, /* Print sorted thread stats (true) */
   GPTLdopr_multparent = #GPTLdopr_multparent, /* Print multiple parent info (true) */
   GPTLdopr_collision  = #GPTLdopr_collision, /* Print hastable collision info (true) */
+  GPTLdopr_memusage   = #GPTLdopr_memusage, /* Call GPTLprint_memusage when auto-instrumented */
   GPTLprint_method    = #GPTLprint_method, /* Tree print method: first parent, last parent
 			       most frequent, or full tree (most frequent) */
   GPTLtablesize       = #GPTLtablesize, /* per-thread size of hash table */

gptl.inc

@@ -22,6 +22,7 @@
       integer GPTLdopr_threadsort
       integer GPTLdopr_multparent
       integer GPTLdopr_collision
+      integer GPTLdopr_memusage
       integer GPTLprint_method
       integer GPTLtablesize
       integer GPTLmaxthreads
@@ -65,6 +66,7 @@
       parameter (GPTLdopr_threadsort= 13)
       parameter (GPTLdopr_multparent= 14)
       parameter (GPTLdopr_collision = 15)
+      parameter (GPTLdopr_memusage  = 27)
       parameter (GPTLprint_method   = 16)
       parameter (GPTLtablesize      = 50)
       parameter (GPTLmaxthreads     = 51)

gptl.inc.template

@@ -22,6 +22,7 @@
       integer GPTLdopr_threadsort
       integer GPTLdopr_multparent
       integer GPTLdopr_collision
+      integer GPTLdopr_memusage
       integer GPTLprint_method
       integer GPTLtablesize
       integer GPTLmaxthreads
@@ -65,6 +66,7 @@
       parameter (GPTLdopr_threadsort= #GPTLdopr_threadsort)
       parameter (GPTLdopr_multparent= #GPTLdopr_multparent)
       parameter (GPTLdopr_collision = #GPTLdopr_collision)
+      parameter (GPTLdopr_memusage  = #GPTLdopr_memusage)
       parameter (GPTLprint_method   = #GPTLprint_method)
       parameter (GPTLtablesize      = #GPTLtablesize)
       parameter (GPTLmaxthreads     = #GPTLmaxthreads)

gptlf.F90

@@ -22,6 +22,7 @@ module gptl
   integer, parameter :: GPTLdopr_threadsort= 13
   integer, parameter :: GPTLdopr_multparent= 14
   integer, parameter :: GPTLdopr_collision = 15
+  integer, parameter :: GPTLdopr_memusage  = 27
   integer, parameter :: GPTLprint_method   = 16
   integer, parameter :: GPTLtablesize      = 50
   integer, parameter :: GPTLmaxthreads     = 51

gptlf.F90.template

@@ -22,6 +22,7 @@ module gptl
   integer, parameter :: GPTLdopr_threadsort= #GPTLdopr_threadsort
   integer, parameter :: GPTLdopr_multparent= #GPTLdopr_multparent
   integer, parameter :: GPTLdopr_collision = #GPTLdopr_collision
+  integer, parameter :: GPTLdopr_memusage  = #GPTLdopr_memusage
   integer, parameter :: GPTLprint_method   = #GPTLprint_method
   integer, parameter :: GPTLtablesize      = #GPTLtablesize
   integer, parameter :: GPTLmaxthreads     = #GPTLmaxthreads

macros.make.jet

@@ -141,3 +141,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = yes
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.jet.nompi

@@ -140,3 +140,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = yes
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.lahey

@@ -141,3 +141,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = no
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.linux

@@ -123,7 +123,7 @@ HAVE_TIMES = yes
 HAVE_GETTIMEOFDAY = yes
 
 # Whether to test auto-profiling (adds 2 tests to "make test"). If compiler is gcc or 
-# pathscale, set INSTRFLAG to -finstrument-functions. PGI 8.0.2 and later provide 
+# pathscale or intel, set INSTRFLAG to -finstrument-functions. PGI 8.0.2 and later provide 
 # -Minstrument:functions.
 TEST_AUTOPROFILE = yes
 ifeq ($(TEST_AUTOPROFILE),yes)
@@ -141,3 +141,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = yes
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.macos

@@ -142,3 +142,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = yes
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.ornlintel

@@ -141,3 +141,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = yes
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif

macros.make.pgi

@@ -141,3 +141,13 @@ ifeq ($(HAVE_MPI),yes)
     HAVE_IARGCGETARG = no
   endif
 endif
+
+# Whether system functions backtrace() and backtrace_symbols() exist. Setting HAVE_BACKTRACE=yes
+# will enable auto-generation of function name when auto-profiling and GPTLdopr_memusage has been
+# enabled at run-time. If HAVE_BACKTRACE=no, function address will be printed instead.
+# GNU compilers: compile application code with -finstrument-functions -rdynamic
+# Intel compilers: compile application code with -finstrument-functions -rdynamic -g
+HAVE_BACKTRACE = no
+ifeq ($(HAVE_BACKTRACE),yes)
+  CFLAGS += -DHAVE_BACKTRACE
+endif