Add funtionality to track the resident set size

Makefile.am

@@ -11,6 +11,7 @@ lib_LTLIBRARIES = libftimings-@FTIMINGS_API_VERSION@-@FC@.la
 libftimings_@FTIMINGS_API_VERSION@_@FC@_la_SOURCES = \
 	ftimings/time.c \
 	ftimings/papi.c \
+	ftimings/resident_set_size.c \
 	ftimings/ftimings_type.F90 \
 	ftimings/ftimings_value.F90 \
 	ftimings/ftimings_node.F90 \

ftimings/ftimings.F90

@@ -305,7 +305,8 @@ module ftimings
     character(len=name_length), parameter :: group = "Group"
     character(len=12), parameter :: seconds = "         [s]"
     character(len=8), parameter :: fract = "fraction"
-    character(len=7), parameter :: flops = "Mflop/s"
+    character(len=8), parameter :: flops = " Mflop/s"
+    character(len=8), parameter :: ram   = "     RSS"
 
     if (.not. self%active) then
       return
@@ -348,17 +349,14 @@ module ftimings
     end if
 
     ! I really do hate it ..
-    write(format_spec,'("("" /= "",a",i0,",2x,a12,5x,a8)")') name_length
-    write(unit_act, format_spec,advance='no') adjustl(group), seconds, fract
+    write(format_spec,'("("" /= "",a",i0,",2x,a12,5x,a8,5x,a8)")') name_length
+    write(unit_act, format_spec,advance='no') adjustl(group), seconds, fract, ram
 #ifdef HAVE_LIBPAPI
     if (papi_supported) then
-      write(unit_act,'(5x,a7)') flops
-    else
-      write(unit_act,'(a)') ""
+      write(unit_act,'(5x,a8)',advance='no') flops
     endif
-#else
-    write(unit_act,'(a)') ""
 #endif
+    write(unit_act,'(a)') ""
 
     call node%print_graph(0, threshold, is_sorted, unit=unit_act)
   end subroutine

ftimings/ftimings_type.F90

 module ftimings_type
-  use, intrinsic :: iso_c_binding, only : C_INT64_T, C_DOUBLE, C_LONG_LONG
+  use, intrinsic :: iso_c_binding, only : C_INT64_T, C_DOUBLE, C_LONG_LONG, C_LONG, C_INT
   implicit none
   integer, parameter :: rk = C_DOUBLE
   integer, parameter :: name_length = 40

ftimings/ftimings_value.F90

@@ -27,8 +27,17 @@ module ftimings_value
   end interface
 #endif
 
+  interface
+    function resident_set_size() result(rsssize) bind(C, name="ftimings_resident_set_size")
+      use, intrinsic :: iso_c_binding
+      implicit none
+      integer(kind=C_LONG) :: rsssize
+    end function
+  end interface
+
   type value_t
     integer(kind=C_INT64_T) :: micros = 0          ! Cumulative microseconds spent in this node
+    integer(kind=C_LONG) :: rsssize = 0
 #ifdef HAVE_LIBPAPI
     integer(kind=C_LONG_LONG) :: flop_count = 0    ! Cumulative floating point operations done in this node
 #endif
@@ -46,7 +55,7 @@ module ftimings_value
   end interface
 
   type(value_t), parameter :: null_value = value_t( &
-                                 micros = 0 &
+                                 micros = 0, rsssize = 0 &
 #ifdef HAVE_LIBPAPI
                                 ,flop_count = 0 &
 #endif
@@ -54,14 +63,17 @@ module ftimings_value
 
   contains
 
-
   ! This is the function that actually returns the current timestamp and all other counters
   function now() result(val)
+    use, intrinsic :: iso_c_binding
     type(value_t) :: val
 
     ! current time
     val%micros = microseconds_since_epoch()
 
+    ! current memory
+    val%rsssize = resident_set_size()
+
 #ifdef HAVE_LIBPAPI
     if (papi_supported) then
       ! flop counter
@@ -77,6 +89,7 @@ module ftimings_value
     class(value_t), intent(in) :: a, b
     type(value_t) :: c
     c%micros = a%micros + b%micros
+    c%rsssize = a%rsssize + b%rsssize
 #ifdef HAVE_LIBPAPI
     c%flop_count = a%flop_count + b%flop_count
 #endif
@@ -86,6 +99,7 @@ module ftimings_value
     class(value_t), intent(in) :: a, b
     type(value_t) :: c
     c%micros = a%micros - b%micros
+    c%rsssize = a%rsssize - b%rsssize
 #ifdef HAVE_LIBPAPI
     c%flop_count = a%flop_count - b%flop_count
 #endif
@@ -95,6 +109,7 @@ module ftimings_value
     class(value_t), intent(in) :: a
     type(value_t) :: neg_a
     neg_a%micros = - a%micros
+    neg_a%rsssize = - a%rsssize
 #ifdef HAVE_LIBPAPI
     neg_a%flop_count = - a%flop_count
 #endif
@@ -108,33 +123,43 @@ module ftimings_value
     character(len=64) :: format_spec
     integer :: unit
 
-    write(format_spec,'("(",i0,"x,""|_ "",a",i0,",2x,f12.6,5x,f8.3)")') indent_level * 2 + 1, name_length
+    write(format_spec,'("(",i0,"x,""|_ "",a",i0,",2x,f12.6,5x,f8.3,5x,a8)")') indent_level * 2 + 1, name_length
     write(unit,format_spec,advance='no') &
-      label, real(self%micros, kind=rk) * 1e-6_rk, real(self%micros, kind=rk) / real(total%micros, kind=rk)
+      label, &
+      real(self%micros, kind=rk) * 1e-6_rk, &
+      real(self%micros, kind=rk) / real(total%micros, kind=rk), &
+      nice_format(real(self%rsssize, kind=C_DOUBLE))
+
 #ifdef HAVE_LIBPAPI
     if (papi_supported) then
-      write(unit,'(5x,f7.2)') real(self%flop_count, kind=C_DOUBLE) / self%micros
-    else
-      write(unit,'(a)') ""
+      write(unit,'(5x,f8.2)',advance='no') real(self%flop_count, kind=C_DOUBLE) / self%micros
     endif
-#else
-    write(unit,'(a)') ""
 #endif
+    write(unit,'(a)') ""
   end subroutine
 
-  pure elemental function format_flops(flops) result(string)
+  pure elemental function nice_format(flops) result(string)
     real(kind=C_DOUBLE), intent(in) :: flops
-    character(len=7) :: string
-    if (flops > 1e12) then
-      write(string,'(f5.1,'' T'')') flops / 1e12
-    else if (flops > 1e9) then
-      write(string,'(f5.1,'' G'')') flops / 1e9
-    else if (flops > 1e6) then
-      write(string,'(f5.1,'' M'')') flops / 1e6
-    else if (flops > 1e3) then
-      write(string,'(f5.1,'' K'')') flops / 1e3
+    character(len=8) :: string
+    real(kind=C_DOUBLE), parameter :: &
+        kibi = 2.0_C_DOUBLE**10, &
+        mebi = 2.0_C_DOUBLE**20, &
+        gibi = 2.0_C_DOUBLE**30, &
+        tebi = 2.0_C_DOUBLE**40, &
+        pebi = 2.0_C_DOUBLE**50
+
+    if (abs(flops) >= pebi) then
+      write(string,'(es8.1)') flops
+    else if (abs(flops) >= tebi) then
+      write(string,'(f5.1,'' Ti'')') flops / tebi
+    else if (abs(flops) >= gibi) then
+      write(string,'(f5.1,'' Gi'')') flops / gibi
+    else if (abs(flops) >= mebi) then
+      write(string,'(f5.1,'' Mi'')') flops / mebi
+    else if (abs(flops) >= kibi) then
+      write(string,'(f5.1,'' ki'')') flops / kibi
     else
-      write(string,'(f7.1)') flops
+      write(string,'(f5.1,''  '')') flops
     endif
   end function
 

ftimings/resident_set_size.c

+#include <stdio.h>
+#include <unistd.h>
+
+/*#include <sys/time.h>
+#include <sys/resource.h>*/
+
+long ftimings_resident_set_size() {
+/*	struct rusage usage;
+	if (getrusage(RUSAGE_SELF, &usage) != 0) {
+		perror("getrusage");
+		exit(1);
+	}
+	return usage.ru_maxrss;
+*/
+
+	long rss = 0L;
+	FILE* fp = NULL;
+	if ((fp = fopen( "/proc/self/statm", "r" )) == NULL ) {
+		return 0L;
+	}
+	if (fscanf(fp, "%*s%ld", &rss) != 1) {
+		fclose(fp);
+		return (size_t)0L;	  /* Can't read? */
+	}
+	fclose(fp);
+	return rss * sysconf( _SC_PAGESIZE);
+}

test/test_timings.F90

@@ -27,7 +27,7 @@ program test_timings
     endif
 
     write(*,*) "Iteration", i
-    call timer%start("cycle", replace=.true.)
+    call timer%start("cycle") ! replace=.true.)
     !$ call timer%start("parallel")
     !$omp parallel do
     do j = 1,30