patch 8.1.1684: profiling functionality is spread outv8.1.1684

Problem:    Profiling functionality is spread out.
Solution:   Put profiling functionality in profiling.c. (Yegappan Lakshmanan,
            closes #4666)

19 files changed, 831 insertions, 755 deletions

diff --git a/src/Make_cyg_ming.mak b/src/Make_cyg_ming.mak
index d0769c39e..b689df8ad 100644
--- a/src/Make_cyg_ming.mak
+++ b/src/Make_cyg_ming.mak
@@ -752,6 +752,7 @@ OBJ = \
 	$(OUTDIR)/pathdef.o \
 	$(OUTDIR)/popupmnu.o \
 	$(OUTDIR)/popupwin.o \
+	$(OUTDIR)/profiler.o \
 	$(OUTDIR)/quickfix.o \
 	$(OUTDIR)/regexp.o \
 	$(OUTDIR)/screen.o \
diff --git a/src/Make_dice.mak b/src/Make_dice.mak
index 440955eb3..f95262ebb 100644
--- a/src/Make_dice.mak
+++ b/src/Make_dice.mak
@@ -71,6 +71,7 @@ SRC = \
 	option.c \
 	os_amiga.c \
 	popupmnu.c \
+	profiler.c \
 	quickfix.c \
 	regexp.c \
 	screen.c \
@@ -134,6 +135,7 @@ OBJ =	o/arabic.o \
 	o/option.o \
 	o/os_amiga.o \
 	o/popupmnu.o \
+	o/profiler.o \
 	o/quickfix.o \
 	o/regexp.o \
 	o/screen.o \
@@ -266,6 +268,8 @@ o/os_amiga.o:	os_amiga.c  $(SYMS) os_amiga.h
 
 o/popupmnu.o:	popupmnu.c  $(SYMS)
 
+o/profiler.o:	profiler.c  $(SYMS)
+
 o/quickfix.o:	quickfix.c  $(SYMS)
 
 o/regexp.o:	regexp.c  $(SYMS) regexp.h
diff --git a/src/Make_manx.mak b/src/Make_manx.mak
index 211d8cceb..e3babe567 100644
--- a/src/Make_manx.mak
+++ b/src/Make_manx.mak
@@ -81,6 +81,7 @@ SRC =	arabic.c \
 	option.c \
 	os_amiga.c \
 	popupmnu.c \
+	profiler.c \
 	quickfix.c \
 	regexp.c \
 	screen.c \
@@ -146,6 +147,7 @@ OBJ =	obj/arabic.o \
 	obj/option.o \
 	obj/os_amiga.o \
 	obj/popupmnu.o \
+	obj/profiler.o \
 	obj/quickfix.o \
 	obj/regexp.o \
 	obj/screen.o \
@@ -209,6 +211,7 @@ PRO =	proto/arabic.pro \
 	proto/option.pro \
 	proto/os_amiga.pro \
 	proto/popupmnu.pro \
+	proto/profiler.pro \
 	proto/quickfix.pro \
 	proto/regexp.pro \
 	proto/screen.pro \
@@ -410,6 +413,9 @@ obj/os_amiga.o:	os_amiga.c
 obj/popupmnu.o:	popupmnu.c
 	$(CCSYM) $@ popupmnu.c
 
+obj/profiler.o:	profiler.c
+	$(CCSYM) $@ profiler.c
+
 obj/quickfix.o:	quickfix.c
 	$(CCSYM) $@ quickfix.c
 
diff --git a/src/Make_morph.mak b/src/Make_morph.mak
index 1eb5ff52d..7bd39faa3 100644
--- a/src/Make_morph.mak
+++ b/src/Make_morph.mak
@@ -69,6 +69,7 @@ SRC =	arabic.c						\
 	option.c						\
 	os_amiga.c						\
 	popupmnu.c						\
+	profiler.c						\
 	quickfix.c						\
 	regexp.c						\
 	screen.c						\
diff --git a/src/Make_mvc.mak b/src/Make_mvc.mak
index 04e566bbb..4a6aaa995 100644
--- a/src/Make_mvc.mak
+++ b/src/Make_mvc.mak
@@ -761,6 +761,7 @@ OBJ = \
 	$(OUTDIR)\pathdef.obj \
 	$(OUTDIR)\popupmnu.obj \
 	$(OUTDIR)\popupwin.obj \
+	$(OUTDIR)\profiler.obj \
 	$(OUTDIR)\quickfix.obj \
 	$(OUTDIR)\regexp.obj \
 	$(OUTDIR)\screen.obj \
@@ -1595,6 +1596,8 @@ $(OUTDIR)/popupmnu.obj:	$(OUTDIR) popupmnu.c  $(INCL)
 
 $(OUTDIR)/popupwin.obj:	$(OUTDIR) popupwin.c  $(INCL)
 
+$(OUTDIR)/profiler.obj:	$(OUTDIR) profiler.c  $(INCL)
+
 $(OUTDIR)/quickfix.obj:	$(OUTDIR) quickfix.c  $(INCL)
 
 $(OUTDIR)/regexp.obj:	$(OUTDIR) regexp.c regexp_nfa.c  $(INCL)
@@ -1763,6 +1766,7 @@ proto.h: \
 	proto/os_win32.pro \
 	proto/popupmnu.pro \
 	proto/popupwin.pro \
+	proto/profiler.pro \
 	proto/quickfix.pro \
 	proto/regexp.pro \
 	proto/screen.pro \
diff --git a/src/Make_sas.mak b/src/Make_sas.mak
index 7090119d7..607f48448 100644
--- a/src/Make_sas.mak
+++ b/src/Make_sas.mak
@@ -134,6 +134,7 @@ SRC = \
 	option.c \
 	os_amiga.c \
 	popupmnu.c \
+	profiler.c \
 	quickfix.c \
 	regexp.c \
 	screen.c \
@@ -198,6 +199,7 @@ OBJ = \
 	option.o \
 	os_amiga.o \
 	popupmnu.o \
+	profiler.o \
 	quickfix.o \
 	regexp.o \
 	screen.o \
@@ -262,6 +264,7 @@ PRO = \
 	proto/option.pro \
 	proto/os_amiga.pro \
 	proto/popupmnu.pro \
+	proto/profiler.pro \
 	proto/quickfix.pro \
 	proto/regexp.pro \
 	proto/screen.pro \
@@ -425,6 +428,8 @@ os_amiga.o:		os_amiga.c
 proto/os_amiga.pro:	os_amiga.c
 popupmnu.o:		popupmnu.c
 proto/popupmnu.pro:	popupmnu.c
+profiler.o:		profiler.c
+proto/profiler.pro:	profiler.c
 quickfix.o:		quickfix.c
 proto/quickfix.pro:	quickfix.c
 regexp.o:		regexp.c
diff --git a/src/Make_vms.mms b/src/Make_vms.mms
index f01d325c0..b7063e158 100644
--- a/src/Make_vms.mms
+++ b/src/Make_vms.mms
@@ -313,10 +313,10 @@ SRC =	arabic.c autocmd.c beval.c blob.c blowfish.c buffer.c change.c charset.c \
 	if_cscope.c if_xcmdsrv.c fileio.c findfile.c fold.c getchar.c \
 	hardcopy.c hashtab.c indent.c insexpand.c json.c list.c main.c mark.c \
 	menu.c mbyte.c memfile.c memline.c message.c misc1.c misc2.c move.c \
-	normal.c ops.c option.c popupmnu.c popupwin.c, quickfix.c regexp.c search.c \
-	sha256.c sign.c spell.c spellfile.c syntax.c tag.c term.c termlib.c \
-	textprop.c ui.c undo.c usercmd.c userfunc.c version.c screen.c \
-	window.c os_unix.c os_vms.c pathdef.c \
+	normal.c ops.c option.c popupmnu.c popupwin.c profiler.c quickfix.c \
+	regexp.c search.c sha256.c sign.c spell.c spellfile.c syntax.c tag.c \
+	term.c termlib.c textprop.c ui.c undo.c usercmd.c userfunc.c \
+	version.c screen.c window.c os_unix.c os_vms.c pathdef.c \
 	$(GUI_SRC) $(PERL_SRC) $(PYTHON_SRC) $(TCL_SRC) \
  	$(RUBY_SRC) $(HANGULIN_SRC) $(MZSCH_SRC) $(XDIFF_SRC)
 
@@ -327,11 +327,12 @@ OBJ = 	arabic.obj autocmd.obj beval.obj blob.obj blowfish.obj buffer.obj change.
 	fileio.obj findfile.obj fold.obj getchar.obj hardcopy.obj hashtab.obj \
 	indent.obj insexpand.obj json.obj list.obj main.obj mark.obj \
 	menu.obj memfile.obj memline.obj message.obj misc1.obj misc2.obj \
-	move.obj mbyte.obj normal.obj ops.obj option.obj popupmnu.obj popupwin.obj\
-	quickfix.obj regexp.obj search.obj sha256.obj sign.obj spell.obj \
-	spellfile.obj syntax.obj tag.obj term.obj termlib.obj textprop.obj \
-	ui.obj undo.obj usercmd.obj userfunc.obj screen.obj version.obj \
-	window.obj os_unix.obj os_vms.obj pathdef.obj if_mzsch.obj \
+	move.obj mbyte.obj normal.obj ops.obj option.obj popupmnu.obj \
+	popupwin.obj profiler.obj quickfix.obj regexp.obj search.obj \
+	sha256.obj sign.obj spell.obj spellfile.obj syntax.obj tag.obj \
+	term.obj termlib.obj textprop.obj ui.obj undo.obj usercmd.obj \
+	userfunc.obj screen.obj version.obj window.obj os_unix.obj os_vms.obj \
+	pathdef.obj if_mzsch.obj \
 	$(GUI_OBJ) $(PERL_OBJ) $(PYTHON_OBJ) $(TCL_OBJ) \
  	$(RUBY_OBJ) $(HANGULIN_OBJ) $(MZSCH_OBJ) $(XDIFF_OBJ)
 
@@ -692,6 +693,10 @@ popupwin.obj : popupwin.c vim.h [.auto]config.h feature.h os_unix.h \
  ascii.h keymap.h term.h macros.h structs.h regexp.h \
  gui.h beval.h [.proto]gui_beval.pro option.h ex_cmds.h proto.h \
  globals.h
+profiler.obj : profiler.c vim.h [.auto]config.h feature.h os_unix.h \
+ ascii.h keymap.h term.h macros.h structs.h regexp.h \
+ gui.h beval.h [.proto]gui_beval.pro option.h ex_cmds.h proto.h \
+ globals.h
 quickfix.obj : quickfix.c vim.h [.auto]config.h feature.h os_unix.h \
  ascii.h keymap.h term.h macros.h structs.h regexp.h \
  gui.h beval.h [.proto]gui_beval.pro option.h ex_cmds.h proto.h \
diff --git a/src/Makefile b/src/Makefile
index 9c4a42893..3f91e5bf1 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -521,6 +521,7 @@ CClink = $(CC)
 # gpm - For mouse support on Linux console via gpm
 # Uncomment this when you do not want to include gpm support, even
 # though you have gpm libraries and includes.
+# For Debian/Ubuntu gpm support requires the libgpm-dev package.
 #CONF_OPT_GPM = --disable-gpm
 
 # sysmouse - For mouse support on FreeBSD and DragonFly console via sysmouse
@@ -1626,6 +1627,7 @@ BASIC_SRC = \
 	auto/pathdef.c \
 	popupmnu.c \
 	popupwin.c \
+	profiler.c \
 	pty.c \
 	quickfix.c \
 	regexp.c \
@@ -1742,6 +1744,7 @@ OBJ_COMMON = \
 	objects/pathdef.o \
 	objects/popupmnu.o \
 	objects/popupwin.o \
+	objects/profiler.o \
 	objects/pty.o \
 	objects/quickfix.o \
 	objects/regexp.o \
@@ -1883,6 +1886,7 @@ PRO_AUTO = \
 	os_unix.pro \
 	popupmnu.pro \
 	popupwin.pro \
+	profiler.pro \
 	pty.pro \
 	quickfix.pro \
 	regexp.pro \
@@ -3222,6 +3226,9 @@ objects/popupmnu.o: popupmnu.c
 objects/popupwin.o: popupwin.c
 	$(CCC) -o $@ popupwin.c
 
+objects/profiler.o: profiler.c
+	$(CCC) -o $@ profiler.c
+
 objects/pty.o: pty.c
 	$(CCC) -o $@ pty.c
 
@@ -3633,6 +3640,10 @@ objects/popupwin.o: popupwin.c vim.h protodef.h auto/config.h feature.h os_unix.
  auto/osdef.h ascii.h keymap.h term.h macros.h option.h beval.h \
  proto/gui_beval.pro structs.h regexp.h gui.h alloc.h ex_cmds.h spell.h \
  proto.h globals.h
+objects/profiler.o: profiler.c vim.h protodef.h auto/config.h feature.h os_unix.h \
+ auto/osdef.h ascii.h keymap.h term.h macros.h option.h beval.h \
+ proto/gui_beval.pro structs.h regexp.h gui.h alloc.h ex_cmds.h spell.h \
+ proto.h globals.h
 objects/pty.o: pty.c vim.h protodef.h auto/config.h feature.h os_unix.h \
  auto/osdef.h ascii.h keymap.h term.h macros.h option.h beval.h \
  proto/gui_beval.pro structs.h regexp.h gui.h alloc.h ex_cmds.h spell.h \
diff --git a/src/README.md b/src/README.md
index ff9f93ac6..87361097f 100644
--- a/src/README.md
+++ b/src/README.md
@@ -44,13 +44,14 @@ menu.c		| menus
 message.c	| (error) messages
 ops.c		| handling operators ("d", "y", "p")
 option.c	| options
+profiler.c	| vim script profiler
 quickfix.c	| quickfix commands (":make", ":cn")
 regexp.c	| pattern matching
 screen.c	| updating the windows
 search.c	| pattern searching
 sign.c		| signs
 spell.c		| spell checking
-syntax.c	|  syntax and other highlighting
+syntax.c	| syntax and other highlighting
 tag.c		| tags
 term.c		| terminal handling, termcap codes
 undo.c		| undo and redo
diff --git a/src/ex_cmds2.c b/src/ex_cmds2.c
index fb26b2aa4..77ff08cf7 100644
--- a/src/ex_cmds2.c
+++ b/src/ex_cmds2.c
@@ -16,215 +16,7 @@
 
 static void	cmd_source(char_u *fname, exarg_T *eap);
 
-#ifdef FEAT_EVAL
-/* Growarray to store info about already sourced scripts.
- * For Unix also store the dev/ino, so that we don't have to stat() each
- * script when going through the list. */
-typedef struct scriptitem_S
-{
-    char_u	*sn_name;
-# ifdef UNIX
-    int		sn_dev_valid;
-    dev_t	sn_dev;
-    ino_t	sn_ino;
-# endif
-# ifdef FEAT_PROFILE
-    int		sn_prof_on;	/* TRUE when script is/was profiled */
-    int		sn_pr_force;	/* forceit: profile functions in this script */
-    proftime_T	sn_pr_child;	/* time set when going into first child */
-    int		sn_pr_nest;	/* nesting for sn_pr_child */
-    /* profiling the script as a whole */
-    int		sn_pr_count;	/* nr of times sourced */
-    proftime_T	sn_pr_total;	/* time spent in script + children */
-    proftime_T	sn_pr_self;	/* time spent in script itself */
-    proftime_T	sn_pr_start;	/* time at script start */
-    proftime_T	sn_pr_children; /* time in children after script start */
-    /* profiling the script per line */
-    garray_T	sn_prl_ga;	/* things stored for every line */
-    proftime_T	sn_prl_start;	/* start time for current line */
-    proftime_T	sn_prl_children; /* time spent in children for this line */
-    proftime_T	sn_prl_wait;	/* wait start time for current line */
-    int		sn_prl_idx;	/* index of line being timed; -1 if none */
-    int		sn_prl_execed;	/* line being timed was executed */
-# endif
-} scriptitem_T;
-
-static garray_T script_items = {0, 0, sizeof(scriptitem_T), 4, NULL};
-#define SCRIPT_ITEM(id) (((scriptitem_T *)script_items.ga_data)[(id) - 1])
-
-# ifdef FEAT_PROFILE
-/* Struct used in sn_prl_ga for every line of a script. */
-typedef struct sn_prl_S
-{
-    int		snp_count;	/* nr of times line was executed */
-    proftime_T	sn_prl_total;	/* time spent in a line + children */
-    proftime_T	sn_prl_self;	/* time spent in a line itself */
-} sn_prl_T;
-
-#  define PRL_ITEM(si, idx)	(((sn_prl_T *)(si)->sn_prl_ga.ga_data)[(idx)])
-# endif
-#endif
-
 #if defined(FEAT_EVAL) || defined(PROTO)
-# if defined(FEAT_PROFILE) || defined(FEAT_RELTIME) || defined(PROTO)
-/*
- * Store the current time in "tm".
- */
-    void
-profile_start(proftime_T *tm)
-{
-# ifdef MSWIN
-    QueryPerformanceCounter(tm);
-# else
-    gettimeofday(tm, NULL);
-# endif
-}
-
-/*
- * Compute the elapsed time from "tm" till now and store in "tm".
- */
-    void
-profile_end(proftime_T *tm)
-{
-    proftime_T now;
-
-# ifdef MSWIN
-    QueryPerformanceCounter(&now);
-    tm->QuadPart = now.QuadPart - tm->QuadPart;
-# else
-    gettimeofday(&now, NULL);
-    tm->tv_usec = now.tv_usec - tm->tv_usec;
-    tm->tv_sec = now.tv_sec - tm->tv_sec;
-    if (tm->tv_usec < 0)
-    {
-	tm->tv_usec += 1000000;
-	--tm->tv_sec;
-    }
-# endif
-}
-
-/*
- * Subtract the time "tm2" from "tm".
- */
-    void
-profile_sub(proftime_T *tm, proftime_T *tm2)
-{
-# ifdef MSWIN
-    tm->QuadPart -= tm2->QuadPart;
-# else
-    tm->tv_usec -= tm2->tv_usec;
-    tm->tv_sec -= tm2->tv_sec;
-    if (tm->tv_usec < 0)
-    {
-	tm->tv_usec += 1000000;
-	--tm->tv_sec;
-    }
-# endif
-}
-
-/*
- * Return a string that represents the time in "tm".
- * Uses a static buffer!
- */
-    char *
-profile_msg(proftime_T *tm)
-{
-    static char buf[50];
-
-# ifdef MSWIN
-    LARGE_INTEGER   fr;
-
-    QueryPerformanceFrequency(&fr);
-    sprintf(buf, "%10.6lf", (double)tm->QuadPart / (double)fr.QuadPart);
-# else
-    sprintf(buf, "%3ld.%06ld", (long)tm->tv_sec, (long)tm->tv_usec);
-# endif
-    return buf;
-}
-
-# if defined(FEAT_FLOAT) || defined(PROTO)
-/*
- * Return a float that represents the time in "tm".
- */
-    float_T
-profile_float(proftime_T *tm)
-{
-#  ifdef MSWIN
-    LARGE_INTEGER   fr;
-
-    QueryPerformanceFrequency(&fr);
-    return (float_T)tm->QuadPart / (float_T)fr.QuadPart;
-#  else
-    return (float_T)tm->tv_sec + (float_T)tm->tv_usec / 1000000.0;
-#  endif
-}
-# endif
-
-/*
- * Put the time "msec" past now in "tm".
- */
-    void
-profile_setlimit(long msec, proftime_T *tm)
-{
-    if (msec <= 0)   /* no limit */
-	profile_zero(tm);
-    else
-    {
-# ifdef MSWIN
-	LARGE_INTEGER   fr;
-
-	QueryPerformanceCounter(tm);
-	QueryPerformanceFrequency(&fr);
-	tm->QuadPart += (LONGLONG)((double)msec / 1000.0 * (double)fr.QuadPart);
-# else
-	long	    usec;
-
-	gettimeofday(tm, NULL);
-	usec = (long)tm->tv_usec + (long)msec * 1000;
-	tm->tv_usec = usec % 1000000L;
-	tm->tv_sec += usec / 1000000L;
-# endif
-    }
-}
-
-/*
- * Return TRUE if the current time is past "tm".
- */
-    int
-profile_passed_limit(proftime_T *tm)
-{
-    proftime_T	now;
-
-# ifdef MSWIN
-    if (tm->QuadPart == 0)  /* timer was not set */
-	return FALSE;
-    QueryPerformanceCounter(&now);
-    return (now.QuadPart > tm->QuadPart);
-# else
-    if (tm->tv_sec == 0)    /* timer was not set */
-	return FALSE;
-    gettimeofday(&now, NULL);
-    return (now.tv_sec > tm->tv_sec
-	    || (now.tv_sec == tm->tv_sec && now.tv_usec > tm->tv_usec));
-# endif
-}
-
-/*
- * Set the time in "tm" to zero.
- */
-    void
-profile_zero(proftime_T *tm)
-{
-# ifdef MSWIN
-    tm->QuadPart = 0;
-# else
-    tm->tv_usec = 0;
-    tm->tv_sec = 0;
-# endif
-}
-
-# endif  /* FEAT_PROFILE || FEAT_RELTIME */
-
 # if defined(FEAT_TIMERS) || defined(PROTO)
 static timer_T	*first_timer = NULL;
 static long	last_timer_id = 0;
@@ -603,270 +395,11 @@ timer_free_all()
 #  endif
 # endif
 
-#if defined(FEAT_SYN_HL) && defined(FEAT_RELTIME) && defined(FEAT_FLOAT) && defined(FEAT_PROFILE)
-# if defined(HAVE_MATH_H)
-#  include <math.h>
-# endif
-
-/*
- * Divide the time "tm" by "count" and store in "tm2".
- */
-    void
-profile_divide(proftime_T *tm, int count, proftime_T *tm2)
-{
-    if (count == 0)
-	profile_zero(tm2);
-    else
-    {
-# ifdef MSWIN
-	tm2->QuadPart = tm->QuadPart / count;
-# else
-	double usec = (tm->tv_sec * 1000000.0 + tm->tv_usec) / count;
-
-	tm2->tv_sec = floor(usec / 1000000.0);
-	tm2->tv_usec = vim_round(usec - (tm2->tv_sec * 1000000.0));
-# endif
-    }
-}
-#endif
-
 # if defined(FEAT_PROFILE) || defined(PROTO)
 /*
- * Functions for profiling.
- */
-static void script_dump_profile(FILE *fd);
-static proftime_T prof_wait_time;
-
-/*
- * Add the time "tm2" to "tm".
- */
-    void
-profile_add(proftime_T *tm, proftime_T *tm2)
-{
-# ifdef MSWIN
-    tm->QuadPart += tm2->QuadPart;
-# else
-    tm->tv_usec += tm2->tv_usec;
-    tm->tv_sec += tm2->tv_sec;
-    if (tm->tv_usec >= 1000000)
-    {
-	tm->tv_usec -= 1000000;
-	++tm->tv_sec;
-    }
-# endif
-}
-
-/*
- * Add the "self" time from the total time and the children's time.
- */
-    void
-profile_self(proftime_T *self, proftime_T *total, proftime_T *children)
-{
-    /* Check that the result won't be negative.  Can happen with recursive
-     * calls. */
-#ifdef MSWIN
-    if (total->QuadPart <= children->QuadPart)
-	return;
-#else
-    if (total->tv_sec < children->tv_sec
-	    || (total->tv_sec == children->tv_sec
-		&& total->tv_usec <= children->tv_usec))
-	return;
-#endif
-    profile_add(self, total);
-    profile_sub(self, children);
-}
-
-/*
- * Get the current waittime.
- */
-    void
-profile_get_wait(proftime_T *tm)
-{
-    *tm = prof_wait_time;
-}
-
-/*
- * Subtract the passed waittime since "tm" from "tma".
- */
-    void
-profile_sub_wait(proftime_T *tm, proftime_T *tma)
-{
-    proftime_T tm3 = prof_wait_time;
-
-    profile_sub(&tm3, tm);
-    profile_sub(tma, &tm3);
-}
-
-/*
- * Return TRUE if "tm1" and "tm2" are equal.
- */
-    int
-profile_equal(proftime_T *tm1, proftime_T *tm2)
-{
-# ifdef MSWIN
-    return (tm1->QuadPart == tm2->QuadPart);
-# else
-    return (tm1->tv_usec == tm2->tv_usec && tm1->tv_sec == tm2->tv_sec);
-# endif
-}
-
-/*
- * Return <0, 0 or >0 if "tm1" < "tm2", "tm1" == "tm2" or "tm1" > "tm2"
- */
-    int
-profile_cmp(const proftime_T *tm1, const proftime_T *tm2)
-{
-# ifdef MSWIN
-    return (int)(tm2->QuadPart - tm1->QuadPart);
-# else
-    if (tm1->tv_sec == tm2->tv_sec)
-	return tm2->tv_usec - tm1->tv_usec;
-    return tm2->tv_sec - tm1->tv_sec;
-# endif
-}
-
-static char_u	*profile_fname = NULL;
-static proftime_T pause_time;
-
-/*
- * ":profile cmd args"
- */
-    void
-ex_profile(exarg_T *eap)
-{
-    char_u	*e;
-    int		len;
-
-    e = skiptowhite(eap->arg);
-    len = (int)(e - eap->arg);
-    e = skipwhite(e);
-
-    if (len == 5 && STRNCMP(eap->arg, "start", 5) == 0 && *e != NUL)
-    {
-	vim_free(profile_fname);
-	profile_fname = expand_env_save_opt(e, TRUE);
-	do_profiling = PROF_YES;
-	profile_zero(&prof_wait_time);
-	set_vim_var_nr(VV_PROFILING, 1L);
-    }
-    else if (do_profiling == PROF_NONE)
-	emsg(_("E750: First use \":profile start {fname}\""));
-    else if (STRCMP(eap->arg, "pause") == 0)
-    {
-	if (do_profiling == PROF_YES)
-	    profile_start(&pause_time);
-	do_profiling = PROF_PAUSED;
-    }
-    else if (STRCMP(eap->arg, "continue") == 0)
-    {
-	if (do_profiling == PROF_PAUSED)
-	{
-	    profile_end(&pause_time);
-	    profile_add(&prof_wait_time, &pause_time);
-	}
-	do_profiling = PROF_YES;
-    }
-    else
-    {
-	/* The rest is similar to ":breakadd". */
-	ex_breakadd(eap);
-    }
-}
-
-/* Command line expansion for :profile. */
-static enum
-{
-    PEXP_SUBCMD,	/* expand :profile sub-commands */
-    PEXP_FUNC		/* expand :profile func {funcname} */
-} pexpand_what;
-
-static char *pexpand_cmds[] = {
-			"start",
-#define PROFCMD_START	0
-			"pause",
-#define PROFCMD_PAUSE	1
-			"continue",
-#define PROFCMD_CONTINUE 2
-			"func",
-#define PROFCMD_FUNC	3
-			"file",
-#define PROFCMD_FILE	4
-			NULL
-#define PROFCMD_LAST	5
-};
-
-/*
- * Function given to ExpandGeneric() to obtain the profile command
- * specific expansion.
- */
-    char_u *
-get_profile_name(expand_T *xp UNUSED, int idx)
-{
-    switch (pexpand_what)
-    {
-    case PEXP_SUBCMD:
-	return (char_u *)pexpand_cmds[idx];
-    /* case PEXP_FUNC: TODO */
-    default:
-	return NULL;
-    }
-}
-
-/*
- * Handle command line completion for :profile command.
- */
-    void
-set_context_in_profile_cmd(expand_T *xp, char_u *arg)
-{
-    char_u	*end_subcmd;
-
-    /* Default: expand subcommands. */
-    xp->xp_context = EXPAND_PROFILE;
-    pexpand_what = PEXP_SUBCMD;
-    xp->xp_pattern = arg;
-
-    end_subcmd = skiptowhite(arg);
-    if (*end_subcmd == NUL)
-	return;
-
-    if (end_subcmd - arg == 5 && STRNCMP(arg, "start", 5) == 0)
-    {
-	xp->xp_context = EXPAND_FILES;
-	xp->xp_pattern = skipwhite(end_subcmd);
-	return;
-    }
-
-    /* TODO: expand function names after "func" */
-    xp->xp_context = EXPAND_NOTHING;
-}
-
-/*
- * Dump the profiling info.
- */
-    void
-profile_dump(void)
-{
-    FILE	*fd;
-
-    if (profile_fname != NULL)
-    {
-	fd = mch_fopen((char *)profile_fname, "w");
-	if (fd == NULL)
-	    semsg(_(e_notopen), profile_fname);
-	else
-	{
-	    script_dump_profile(fd);
-	    func_dump_profile(fd);
-	    fclose(fd);
-	}
-    }
-}
-
-/*
  * Start profiling script "fp".
  */
-    static void
+    void
 script_do_profile(scriptitem_T *si)
 {
     si->sn_pr_count = 0;
@@ -918,31 +451,10 @@ script_prof_restore(proftime_T *tm)
     }
 }
 
-static proftime_T inchar_time;
-
-/*
- * Called when starting to wait for the user to type a character.
- */
-    void
-prof_inchar_enter(void)
-{
-    profile_start(&inchar_time);
-}
-
-/*
- * Called when finished waiting for the user to type a character.
- */
-    void
-prof_inchar_exit(void)
-{
-    profile_end(&inchar_time);
-    profile_add(&prof_wait_time, &inchar_time);
-}
-
 /*
  * Dump the profiling results for all scripts in file "fd".
  */
-    static void
+    void
 script_dump_profile(FILE *fd)
 {
     int		    id;
@@ -1016,19 +528,6 @@ script_dump_profile(FILE *fd)
 	}
     }
 }
-
-/*
- * Return TRUE when a function defined in the current script should be
- * profiled.
- */
-    int
-prof_def_func(void)
-{
-    if (current_sctx.sc_sid > 0)
-	return SCRIPT_ITEM(current_sctx.sc_sid).sn_pr_force;
-    return FALSE;
-}
-
 # endif
 #endif
 
diff --git a/src/globals.h b/src/globals.h
index 3562b5ee6..0c2df36b8 100644
--- a/src/globals.h
+++ b/src/globals.h
@@ -252,6 +252,9 @@ EXTERN int	debug_backtrace_level INIT(= 0); /* breakpoint backtrace level */
 # ifdef FEAT_PROFILE
 EXTERN int	do_profiling INIT(= PROF_NONE);	/* PROF_ values */
 # endif
+EXTERN garray_T script_items INIT(= {0 COMMA 0 COMMA sizeof(scriptitem_T) COMMA 4 COMMA NULL});
+#define SCRIPT_ITEM(id) (((scriptitem_T *)script_items.ga_data)[(id) - 1])
+#define FUNCLINE(fp, j)	((char_u **)(fp->uf_lines.ga_data))[j]
 
 /*
  * The exception currently being thrown.  Used to pass an exception to
diff --git a/src/profiler.c b/src/profiler.c
new file mode 100644
index 000000000..dab21bceb
--- /dev/null
+++ b/src/profiler.c
@@ -0,0 +1,679 @@
+/* vi:set ts=8 sts=4 sw=4 noet:
+ *
+ * VIM - Vi IMproved	by Bram Moolenaar
+ *
+ * Do ":help uganda"  in Vim to read copying and usage conditions.
+ * Do ":help credits" in Vim to see a list of people who contributed.
+ * See README.txt for an overview of the Vim source code.
+ */
+
+/*
+ * profiler.c: vim script profiler
+ */
+
+#include "vim.h"
+
+#if defined(FEAT_EVAL) || defined(PROTO)
+# if defined(FEAT_PROFILE) || defined(FEAT_RELTIME) || defined(PROTO)
+/*
+ * Store the current time in "tm".
+ */
+    void
+profile_start(proftime_T *tm)
+{
+# ifdef MSWIN
+    QueryPerformanceCounter(tm);
+# else
+    gettimeofday(tm, NULL);
+# endif
+}
+
+/*
+ * Compute the elapsed time from "tm" till now and store in "tm".
+ */
+    void
+profile_end(proftime_T *tm)
+{
+    proftime_T now;
+
+# ifdef MSWIN
+    QueryPerformanceCounter(&now);
+    tm->QuadPart = now.QuadPart - tm->QuadPart;
+# else
+    gettimeofday(&now, NULL);
+    tm->tv_usec = now.tv_usec - tm->tv_usec;
+    tm->tv_sec = now.tv_sec - tm->tv_sec;
+    if (tm->tv_usec < 0)
+    {
+	tm->tv_usec += 1000000;
+	--tm->tv_sec;
+    }
+# endif
+}
+
+/*
+ * Subtract the time "tm2" from "tm".
+ */
+    void
+profile_sub(proftime_T *tm, proftime_T *tm2)
+{
+# ifdef MSWIN
+    tm->QuadPart -= tm2->QuadPart;
+# else
+    tm->tv_usec -= tm2->tv_usec;
+    tm->tv_sec -= tm2->tv_sec;
+    if (tm->tv_usec < 0)
+    {
+	tm->tv_usec += 1000000;
+	--tm->tv_sec;
+    }
+# endif
+}
+
+/*
+ * Return a string that represents the time in "tm".
+ * Uses a static buffer!
+ */
+    char *
+profile_msg(proftime_T *tm)
+{
+    static char buf[50];
+
+# ifdef MSWIN
+    LARGE_INTEGER   fr;
+
+    QueryPerformanceFrequency(&fr);
+    sprintf(buf, "%10.6lf", (double)tm->QuadPart / (double)fr.QuadPart);
+# else
+    sprintf(buf, "%3ld.%06ld", (long)tm->tv_sec, (long)tm->tv_usec);
+# endif
+    return buf;
+}
+
+# if defined(FEAT_FLOAT) || defined(PROTO)
+/*
+ * Return a float that represents the time in "tm".
+ */
+    float_T
+profile_float(proftime_T *tm)
+{
+#  ifdef MSWIN
+    LARGE_INTEGER   fr;
+
+    QueryPerformanceFrequency(&fr);
+    return (float_T)tm->QuadPart / (float_T)fr.QuadPart;
+#  else
+    return (float_T)tm->tv_sec + (float_T)tm->tv_usec / 1000000.0;
+#  endif
+}
+# endif
+
+/*
+ * Put the time "msec" past now in "tm".
+ */
+    void
+profile_setlimit(long msec, proftime_T *tm)
+{
+    if (msec <= 0)   /* no limit */
+	profile_zero(tm);
+    else
+    {
+# ifdef MSWIN
+	LARGE_INTEGER   fr;
+
+	QueryPerformanceCounter(tm);
+	QueryPerformanceFrequency(&fr);
+	tm->QuadPart += (LONGLONG)((double)msec / 1000.0 * (double)fr.QuadPart);
+# else
+	long	    usec;
+
+	gettimeofday(tm, NULL);
+	usec = (long)tm->tv_usec + (long)msec * 1000;
+	tm->tv_usec = usec % 1000000L;
+	tm->tv_sec += usec / 1000000L;
+# endif
+    }
+}
+
+/*
+ * Return TRUE if the current time is past "tm".
+ */
+    int
+profile_passed_limit(proftime_T *tm)
+{
+    proftime_T	now;
+
+# ifdef MSWIN
+    if (tm->QuadPart == 0)  /* timer was not set */
+	return FALSE;
+    QueryPerformanceCounter(&now);
+    return (now.QuadPart > tm->QuadPart);
+# else
+    if (tm->tv_sec == 0)    /* timer was not set */
+	return FALSE;
+    gettimeofday(&now, NULL);
+    return (now.tv_sec > tm->tv_sec
+	    || (now.tv_sec == tm->tv_sec && now.tv_usec > tm->tv_usec));
+# endif
+}
+
+/*
+ * Set the time in "tm" to zero.
+ */
+    void
+profile_zero(proftime_T *tm)
+{
+# ifdef MSWIN
+    tm->QuadPart = 0;
+# else
+    tm->tv_usec = 0;
+    tm->tv_sec = 0;
+# endif
+}
+
+# endif  /* FEAT_PROFILE || FEAT_RELTIME */
+
+#if defined(FEAT_SYN_HL) && defined(FEAT_RELTIME) && defined(FEAT_FLOAT) && defined(FEAT_PROFILE)
+# if defined(HAVE_MATH_H)
+#  include <math.h>
+# endif
+
+/*
+ * Divide the time "tm" by "count" and store in "tm2".
+ */
+    void
+profile_divide(proftime_T *tm, int count, proftime_T *tm2)
+{
+    if (count == 0)
+	profile_zero(tm2);
+    else
+    {
+# ifdef MSWIN
+	tm2->QuadPart = tm->QuadPart / count;
+# else
+	double usec = (tm->tv_sec * 1000000.0 + tm->tv_usec) / count;
+
+	tm2->tv_sec = floor(usec / 1000000.0);
+	tm2->tv_usec = vim_round(usec - (tm2->tv_sec * 1000000.0));
+# endif
+    }
+}
+#endif
+
+# if defined(FEAT_PROFILE) || defined(PROTO)
+/*
+ * Functions for profiling.
+ */
+static proftime_T prof_wait_time;
+
+/*
+ * Add the time "tm2" to "tm".
+ */
+    void
+profile_add(proftime_T *tm, proftime_T *tm2)
+{
+# ifdef MSWIN
+    tm->QuadPart += tm2->QuadPart;
+# else
+    tm->tv_usec += tm2->tv_usec;
+    tm->tv_sec += tm2->tv_sec;
+    if (tm->tv_usec >= 1000000)
+    {
+	tm->tv_usec -= 1000000;
+	++tm->tv_sec;
+    }
+# endif
+}
+
+/*
+ * Add the "self" time from the total time and the children's time.
+ */
+    void
+profile_self(proftime_T *self, proftime_T *total, proftime_T *children)
+{
+    /* Check that the result won't be negative.  Can happen with recursive
+     * calls. */
+#ifdef MSWIN
+    if (total->QuadPart <= children->QuadPart)
+	return;
+#else
+    if (total->tv_sec < children->tv_sec
+	    || (total->tv_sec == children->tv_sec
+		&& total->tv_usec <= children->tv_usec))
+	return;
+#endif
+    profile_add(self, total);
+    profile_sub(self, children);
+}
+
+/*
+ * Get the current waittime.
+ */
+    void
+profile_get_wait(proftime_T *tm)
+{
+    *tm = prof_wait_time;
+}
+
+/*
+ * Subtract the passed waittime since "tm" from "tma".
+ */
+    void
+profile_sub_wait(proftime_T *tm, proftime_T *tma)
+{
+    proftime_T tm3 = prof_wait_time;
+
+    profile_sub(&tm3, tm);
+    profile_sub(tma, &tm3);
+}
+
+/*
+ * Return TRUE if "tm1" and "tm2" are equal.
+ */
+    int
+profile_equal(proftime_T *tm1, proftime_T *tm2)
+{
+# ifdef MSWIN
+    return (tm1->QuadPart == tm2->QuadPart);
+# else
+    return (tm1->tv_usec == tm2->tv_usec && tm1->tv_sec == tm2->tv_sec);
+# endif
+}
+
+/*
+ * Return <0, 0 or >0 if "tm1" < "tm2", "tm1" == "tm2" or "tm1" > "tm2"
+ */
+    int
+profile_cmp(const proftime_T *tm1, const proftime_T *tm2)
+{
+# ifdef MSWIN
+    return (int)(tm2->QuadPart - tm1->QuadPart);
+# else
+    if (tm1->tv_sec == tm2->tv_sec)
+	return tm2->tv_usec - tm1->tv_usec;
+    return tm2->tv_sec - tm1->tv_sec;
+# endif
+}
+
+static char_u	*profile_fname = NULL;
+static proftime_T pause_time;
+
+/*
+ * ":profile cmd args"
+ */
+    void
+ex_profile(exarg_T *eap)
+{
+    char_u	*e;
+    int		len;
+
+    e = skiptowhite(eap->arg);
+    len = (int)(e - eap->arg);
+    e = skipwhite(e);
+
+    if (len == 5 && STRNCMP(eap->arg, "start", 5) == 0 && *e != NUL)
+    {
+	vim_free(profile_fname);
+	profile_fname = expand_env_save_opt(e, TRUE);
+	do_profiling = PROF_YES;
+	profile_zero(&prof_wait_time);
+	set_vim_var_nr(VV_PROFILING, 1L);
+    }
+    else if (do_profiling == PROF_NONE)
+	emsg(_("E750: First use \":profile start {fname}\""));
+    else if (STRCMP(eap->arg, "pause") == 0)
+    {
+	if (do_profiling == PROF_YES)
+	    profile_start(&pause_time);
+	do_profiling = PROF_PAUSED;
+    }
+    else if (STRCMP(eap->arg, "continue") == 0)
+    {
+	if (do_profiling == PROF_PAUSED)
+	{
+	    profile_end(&pause_time);
+	    profile_add(&prof_wait_time, &pause_time);
+	}
+	do_profiling = PROF_YES;
+    }
+    else
+    {
+	/* The rest is similar to ":breakadd". */
+	ex_breakadd(eap);
+    }
+}
+
+/* Command line expansion for :profile. */
+static enum
+{
+    PEXP_SUBCMD,	/* expand :profile sub-commands */
+    PEXP_FUNC		/* expand :profile func {funcname} */
+} pexpand_what;
+
+static char *pexpand_cmds[] = {
+			"start",
+#define PROFCMD_START	0
+			"pause",
+#define PROFCMD_PAUSE	1
+			"continue",
+#define PROFCMD_CONTINUE 2
+			"func",
+#define PROFCMD_FUNC	3
+			"file",
+#define PROFCMD_FILE	4
+			NULL
+#define PROFCMD_LAST	5
+};
+
+/*
+ * Function given to ExpandGeneric() to obtain the profile command
+ * specific expansion.
+ */
+    char_u *
+get_profile_name(expand_T *xp UNUSED, int idx)
+{
+    switch (pexpand_what)
+    {
+    case PEXP_SUBCMD:
+	return (char_u *)pexpand_cmds[idx];
+    /* case PEXP_FUNC: TODO */
+    default:
+	return NULL;
+    }
+}
+
+/*
+ * Handle command line completion for :profile command.
+ */
+    void
+set_context_in_profile_cmd(expand_T *xp, char_u *arg)
+{
+    char_u	*end_subcmd;
+
+    /* Default: expand subcommands. */
+    xp->xp_context = EXPAND_PROFILE;
+    pexpand_what = PEXP_SUBCMD;
+    xp->xp_pattern = arg;
+
+    end_subcmd = skiptowhite(arg);
+    if (*end_subcmd == NUL)
+	return;
+
+    if (end_subcmd - arg == 5 && STRNCMP(arg, "start", 5) == 0)
+    {
+	xp->xp_context = EXPAND_FILES;
+	xp->xp_pattern = skipwhite(end_subcmd);
+	return;
+    }
+
+    /* TODO: expand function names after "func" */
+    xp->xp_context = EXPAND_NOTHING;
+}
+
+/*
+ * Dump the profiling info.
+ */
+    void
+profile_dump(void)
+{
+    FILE	*fd;
+
+    if (profile_fname != NULL)
+    {
+	fd = mch_fopen((char *)profile_fname, "w");
+	if (fd == NULL)
+	    semsg(_(e_notopen), profile_fname);
+	else
+	{
+	    script_dump_profile(fd);
+	    func_dump_profile(fd);
+	    fclose(fd);
+	}
+    }
+}
+
+static proftime_T inchar_time;
+
+/*
+ * Called when starting to wait for the user to type a character.
+ */
+    void
+prof_inchar_enter(void)
+{
+    profile_start(&inchar_time);
+}
+
+/*
+ * Called when finished waiting for the user to type a character.
+ */
+    void
+prof_inchar_exit(void)
+{
+    profile_end(&inchar_time);
+    profile_add(&prof_wait_time, &inchar_time);
+}
+
+
+/*
+ * Return TRUE when a function defined in the current script should be
+ * profiled.
+ */
+    int
+prof_def_func(void)
+{
+    if (current_sctx.sc_sid > 0)
+	return SCRIPT_ITEM(current_sctx.sc_sid).sn_pr_force;
+    return FALSE;
+}
+
+    void
+prof_sort_list(
+    FILE	*fd,
+    ufunc_T	**sorttab,
+    int		st_len,
+    char	*title,
+    int		prefer_self)	/* when equal print only self time */
+{
+    int		i;
+    ufunc_T	*fp;
+
+    fprintf(fd, "FUNCTIONS SORTED ON %s TIME\n", title);
+    fprintf(fd, "count  total (s)   self (s)  function\n");
+    for (i = 0; i < 20 && i < st_len; ++i)
+    {
+	fp = sorttab[i];
+	prof_func_line(fd, fp->uf_tm_count, &fp->uf_tm_total, &fp->uf_tm_self,
+								 prefer_self);
+	if (fp->uf_name[0] == K_SPECIAL)
+	    fprintf(fd, " <SNR>%s()\n", fp->uf_name + 3);
+	else
+	    fprintf(fd, " %s()\n", fp->uf_name);
+    }
+    fprintf(fd, "\n");
+}
+
+/*
+ * Print the count and times for one function or function line.
+ */
+    void
+prof_func_line(
+    FILE	*fd,
+    int		count,
+    proftime_T	*total,
+    proftime_T	*self,
+    int		prefer_self)	/* when equal print only self time */
+{
+    if (count > 0)
+    {
+	fprintf(fd, "%5d ", count);
+	if (prefer_self && profile_equal(total, self))
+	    fprintf(fd, "           ");
+	else
+	    fprintf(fd, "%s ", profile_msg(total));
+	if (!prefer_self && profile_equal(total, self))
+	    fprintf(fd, "           ");
+	else
+	    fprintf(fd, "%s ", profile_msg(self));
+    }
+    else
+	fprintf(fd, "                            ");
+}
+
+/*
+ * Compare function for total time sorting.
+ */
+    int
+prof_total_cmp(const void *s1, const void *s2)
+{
+    ufunc_T	*p1, *p2;
+
+    p1 = *(ufunc_T **)s1;
+    p2 = *(ufunc_T **)s2;
+    return profile_cmp(&p1->uf_tm_total, &p2->uf_tm_total);
+}
+
+/*
+ * Compare function for self time sorting.
+ */
+    int
+prof_self_cmp(const void *s1, const void *s2)
+{
+    ufunc_T	*p1, *p2;
+
+    p1 = *(ufunc_T **)s1;
+    p2 = *(ufunc_T **)s2;
+    return profile_cmp(&p1->uf_tm_self, &p2->uf_tm_self);
+}
+
+/*
+ * Start profiling function "fp".
+ */
+    void
+func_do_profile(ufunc_T *fp)
+{
+    int		len = fp->uf_lines.ga_len;
+
+    if (!fp->uf_prof_initialized)
+    {
+	if (len == 0)
+	    len = 1;  /* avoid getting error for allocating zero bytes */
+	fp->uf_tm_count = 0;
+	profile_zero(&fp->uf_tm_self);
+	profile_zero(&fp->uf_tm_total);
+	if (fp->uf_tml_count == NULL)
+	    fp->uf_tml_count = ALLOC_CLEAR_MULT(int, len);
+	if (fp->uf_tml_total == NULL)
+	    fp->uf_tml_total = ALLOC_CLEAR_MULT(proftime_T, len);
+	if (fp->uf_tml_self == NULL)
+	    fp->uf_tml_self = ALLOC_CLEAR_MULT(proftime_T, len);
+	fp->uf_tml_idx = -1;
+	if (fp->uf_tml_count == NULL || fp->uf_tml_total == NULL
+						    || fp->uf_tml_self == NULL)
+	    return;	    /* out of memory */
+	fp->uf_prof_initialized = TRUE;
+    }
+
+    fp->uf_profiling = TRUE;
+}
+
+/*
+ * Prepare profiling for entering a child or something else that is not
+ * counted for the script/function itself.
+ * Should always be called in pair with prof_child_exit().
+ */
+    void
+prof_child_enter(
+    proftime_T *tm)	/* place to store waittime */
+{
+    funccall_T *fc = get_current_funccal();
+
+    if (fc != NULL && fc->func->uf_profiling)
+	profile_start(&fc->prof_child);
+    script_prof_save(tm);
+}
+
+/*
+ * Take care of time spent in a child.
+ * Should always be called after prof_child_enter().
+ */
+    void
+prof_child_exit(
+    proftime_T *tm)	/* where waittime was stored */
+{
+    funccall_T *fc = get_current_funccal();
+
+    if (fc != NULL && fc->func->uf_profiling)
+    {
+	profile_end(&fc->prof_child);
+	profile_sub_wait(tm, &fc->prof_child); /* don't count waiting time */
+	profile_add(&fc->func->uf_tm_children, &fc->prof_child);
+	profile_add(&fc->func->uf_tml_children, &fc->prof_child);
+    }
+    script_prof_restore(tm);
+}
+
+/*
+ * Called when starting to read a function line.
+ * "sourcing_lnum" must be correct!
+ * When skipping lines it may not actually be executed, but we won't find out
+ * until later and we need to store the time now.
+ */
+    void
+func_line_start(void *cookie)
+{
+    funccall_T	*fcp = (funccall_T *)cookie;
+    ufunc_T	*fp = fcp->func;
+
+    if (fp->uf_profiling && sourcing_lnum >= 1
+				      && sourcing_lnum <= fp->uf_lines.ga_len)
+    {
+	fp->uf_tml_idx = sourcing_lnum - 1;
+	/* Skip continuation lines. */
+	while (fp->uf_tml_idx > 0 && FUNCLINE(fp, fp->uf_tml_idx) == NULL)
+	    --fp->uf_tml_idx;
+	fp->uf_tml_execed = FALSE;
+	profile_start(&fp->uf_tml_start);
+	profile_zero(&fp->uf_tml_children);
+	profile_get_wait(&fp->uf_tml_wait);
+    }
+}
+
+/*
+ * Called when actually executing a function line.
+ */
+    void
+func_line_exec(void *cookie)
+{
+    funccall_T	*fcp = (funccall_T *)cookie;
+    ufunc_T	*fp = fcp->func;
+
+    if (fp->uf_profiling && fp->uf_tml_idx >= 0)
+	fp->uf_tml_execed = TRUE;
+}
+
+/*
+ * Called when done with a function line.
+ */
+    void
+func_line_end(void *cookie)
+{
+    funccall_T	*fcp = (funccall_T *)cookie;
+    ufunc_T	*fp = fcp->func;
+
+    if (fp->uf_profiling && fp->uf_tml_idx >= 0)
+    {
+	if (fp->uf_tml_execed)
+	{
+	    ++fp->uf_tml_count[fp->uf_tml_idx];
+	    profile_end(&fp->uf_tml_start);
+	    profile_sub_wait(&fp->uf_tml_wait, &fp->uf_tml_start);
+	    profile_add(&fp->uf_tml_total[fp->uf_tml_idx], &fp->uf_tml_start);
+	    profile_self(&fp->uf_tml_self[fp->uf_tml_idx], &fp->uf_tml_start,
+							&fp->uf_tml_children);
+	}
+	fp->uf_tml_idx = -1;
+    }
+}
+# endif /* FEAT_PROFILE */
+
+#endif
diff --git a/src/proto.h b/src/proto.h
index 1c37bfa18..9119a528c 100644
--- a/src/proto.h
+++ b/src/proto.h
@@ -171,6 +171,9 @@ void qsort(void *base, size_t elm_count, size_t elm_size, int (*cmp)(const void
 # include "ops.pro"
 # include "option.pro"
 # include "popupmnu.pro"
+# if defined(FEAT_PROFILE) || defined(FEAT_RELTIME)
+# include "profiler.pro"
+# endif
 # ifdef FEAT_QUICKFIX
 #  include "quickfix.pro"
 # endif
diff --git a/src/proto/ex_cmds2.pro b/src/proto/ex_cmds2.pro
index c12e0d730..ee74b0b49 100644
--- a/src/proto/ex_cmds2.pro
+++ b/src/proto/ex_cmds2.pro
@@ -1,12 +1,4 @@
 /* ex_cmds2.c */
-void profile_start(proftime_T *tm);
-void profile_end(proftime_T *tm);
-void profile_sub(proftime_T *tm, proftime_T *tm2);
-char *profile_msg(proftime_T *tm);
-float_T profile_float(proftime_T *tm);
-void profile_setlimit(long msec, proftime_T *tm);
-int profile_passed_limit(proftime_T *tm);
-void profile_zero(proftime_T *tm);
 long proftime_time_left(proftime_T *due, proftime_T *now);
 timer_T *create_timer(long msec, int repeat);
 long check_due_timer(void);
@@ -17,22 +9,10 @@ void add_timer_info(typval_T *rettv, timer_T *timer);
 void add_timer_info_all(typval_T *rettv);
 int set_ref_in_timer(int copyID);
 void timer_free_all(void);
-void profile_divide(proftime_T *tm, int count, proftime_T *tm2);
-void profile_add(proftime_T *tm, proftime_T *tm2);
-void profile_self(proftime_T *self, proftime_T *total, proftime_T *children);
-void profile_get_wait(proftime_T *tm);
-void profile_sub_wait(proftime_T *tm, proftime_T *tma);
-int profile_equal(proftime_T *tm1, proftime_T *tm2);
-int profile_cmp(const proftime_T *tm1, const proftime_T *tm2);
-void ex_profile(exarg_T *eap);
-char_u *get_profile_name(expand_T *xp, int idx);
-void set_context_in_profile_cmd(expand_T *xp, char_u *arg);
-void profile_dump(void);
+void script_do_profile(scriptitem_T *si);
 void script_prof_save(proftime_T *tm);
 void script_prof_restore(proftime_T *tm);
-void prof_inchar_enter(void);
-void prof_inchar_exit(void);
-int prof_def_func(void);
+void script_dump_profile(FILE *fd);
 int autowrite(buf_T *buf, int forceit);
 void autowrite_all(void);
 int check_changed(buf_T *buf, int flags);
diff --git a/src/proto/profiler.pro b/src/proto/profiler.pro
new file mode 100644
index 000000000..e7bcb479a
--- /dev/null
+++ b/src/proto/profiler.pro
@@ -0,0 +1,34 @@
+/* profiler.c */
+void profile_start(proftime_T *tm);
+void profile_end(proftime_T *tm);
+void profile_sub(proftime_T *tm, proftime_T *tm2);
+char *profile_msg(proftime_T *tm);
+float_T profile_float(proftime_T *tm);
+void profile_setlimit(long msec, proftime_T *tm);
+int profile_passed_limit(proftime_T *tm);
+void profile_zero(proftime_T *tm);
+void profile_divide(proftime_T *tm, int count, proftime_T *tm2);
+void profile_add(proftime_T *tm, proftime_T *tm2);
+void profile_self(proftime_T *self, proftime_T *total, proftime_T *children);
+void profile_get_wait(proftime_T *tm);
+void profile_sub_wait(proftime_T *tm, proftime_T *tma);
+int profile_equal(proftime_T *tm1, proftime_T *tm2);
+int profile_cmp(const proftime_T *tm1, const proftime_T *tm2);
+void ex_profile(exarg_T *eap);
+char_u *get_profile_name(expand_T *xp, int idx);
+void set_context_in_profile_cmd(expand_T *xp, char_u *arg);
+void profile_dump(void);
+void prof_inchar_enter(void);
+void prof_inchar_exit(void);
+int prof_def_func(void);
+void prof_sort_list(FILE *fd, ufunc_T **sorttab, int st_len, char *title, int prefer_self);
+void prof_func_line(FILE *fd, int count, proftime_T *total, proftime_T *self, int prefer_self);
+int prof_total_cmp(const void *s1, const void *s2);
+int prof_self_cmp(const void *s1, const void *s2);
+void func_do_profile(ufunc_T *fp);
+void prof_child_enter(proftime_T *tm);
+void prof_child_exit(proftime_T *tm);
+void func_line_start(void *cookie);
+void func_line_exec(void *cookie);
+void func_line_end(void *cookie);
+/* vim: set ft=c : */
diff --git a/src/proto/userfunc.pro b/src/proto/userfunc.pro
index c9dd974fb..8395e01b9 100644
--- a/src/proto/userfunc.pro
+++ b/src/proto/userfunc.pro
@@ -6,6 +6,7 @@ int get_func_tv(char_u *name, int len, typval_T *rettv, char_u **arg, linenr_T f
 ufunc_T *find_func(char_u *name);
 void save_funccal(funccal_entry_T *entry);
 void restore_funccal(void);
+funccall_T * get_current_funccal(void);
 void free_all_functions(void);
 int func_call(char_u *name, typval_T *args, partial_T *partial, dict_T *selfdict, typval_T *rettv);
 int call_callback(callback_T *callback, int len, typval_T *rettv, int argcount, typval_T *argvars, int (*argv_func)(int, typval_T *, int), linenr_T firstline, linenr_T lastline, int *doesrange, int evaluate, dict_T *selfdict);
@@ -17,8 +18,6 @@ int translated_function_exists(char_u *name);
 int function_exists(char_u *name, int no_deref);
 char_u *get_expanded_name(char_u *name, int check);
 void func_dump_profile(FILE *fd);
-void prof_child_enter(proftime_T *tm);
-void prof_child_exit(proftime_T *tm);
 char_u *get_user_func_name(expand_T *xp, int idx);
 void ex_delfunction(exarg_T *eap);
 void func_unref(char_u *name);
diff --git a/src/structs.h b/src/structs.h
index 57a09fa98..0ab20354d 100644
--- a/src/structs.h
+++ b/src/structs.h
@@ -1518,6 +1518,49 @@ struct funccal_entry {
     funccal_entry_T *next;
 };
 
+/* Growarray to store info about already sourced scripts.
+ * For Unix also store the dev/ino, so that we don't have to stat() each
+ * script when going through the list. */
+typedef struct scriptitem_S
+{
+    char_u	*sn_name;
+# ifdef UNIX
+    int		sn_dev_valid;
+    dev_t	sn_dev;
+    ino_t	sn_ino;
+# endif
+# ifdef FEAT_PROFILE
+    int		sn_prof_on;	/* TRUE when script is/was profiled */
+    int		sn_pr_force;	/* forceit: profile functions in this script */
+    proftime_T	sn_pr_child;	/* time set when going into first child */
+    int		sn_pr_nest;	/* nesting for sn_pr_child */
+    /* profiling the script as a whole */
+    int		sn_pr_count;	/* nr of times sourced */
+    proftime_T	sn_pr_total;	/* time spent in script + children */
+    proftime_T	sn_pr_self;	/* time spent in script itself */
+    proftime_T	sn_pr_start;	/* time at script start */
+    proftime_T	sn_pr_children; /* time in children after script start */
+    /* profiling the script per line */
+    garray_T	sn_prl_ga;	/* things stored for every line */
+    proftime_T	sn_prl_start;	/* start time for current line */
+    proftime_T	sn_prl_children; /* time spent in children for this line */
+    proftime_T	sn_prl_wait;	/* wait start time for current line */
+    int		sn_prl_idx;	/* index of line being timed; -1 if none */
+    int		sn_prl_execed;	/* line being timed was executed */
+# endif
+} scriptitem_T;
+
+# ifdef FEAT_PROFILE
+/* Struct used in sn_prl_ga for every line of a script. */
+typedef struct sn_prl_S
+{
+    int		snp_count;	/* nr of times line was executed */
+    proftime_T	sn_prl_total;	/* time spent in a line + children */
+    proftime_T	sn_prl_self;	/* time spent in a line itself */
+} sn_prl_T;
+
+#  define PRL_ITEM(si, idx)	(((sn_prl_T *)(si)->sn_prl_ga.ga_data)[(idx)])
+# endif
 #else
 // dummy typedefs for use in function prototypes
 typedef struct
@@ -1527,11 +1570,19 @@ typedef struct
 typedef struct
 {
     int	    dummy;
+} funccall_T;
+typedef struct
+{
+    int	    dummy;
 } funcdict_T;
 typedef struct
 {
     int	    dummy;
 } funccal_entry_T;
+typedef struct
+{
+    int	    dummy;
+} scriptitem_T;
 #endif
 
 struct partial_S
diff --git a/src/userfunc.c b/src/userfunc.c
index 6bfe534a9..6d21b3fbb 100644
--- a/src/userfunc.c
+++ b/src/userfunc.c
@@ -29,7 +29,6 @@
 #define HI2UF(hi)     HIKEY2UF((hi)->hi_key)
 
 #define FUNCARG(fp, j)	((char_u **)(fp->uf_args.ga_data))[j]
-#define FUNCLINE(fp, j)	((char_u **)(fp->uf_lines.ga_data))[j]
 
 /*
  * All user-defined functions are found in this hashtable.
@@ -51,13 +50,6 @@ static char *e_funcdict = N_("E717: Dictionary entry already exists");
 static char *e_funcref = N_("E718: Funcref required");
 static char *e_nofunc = N_("E130: Unknown function: %s");
 
-#ifdef FEAT_PROFILE
-static void func_do_profile(ufunc_T *fp);
-static void prof_sort_list(FILE *fd, ufunc_T **sorttab, int st_len, char *title, int prefer_self);
-static void prof_func_line(FILE *fd, int count, proftime_T *total, proftime_T *self, int prefer_self);
-static int prof_total_cmp(const void *s1, const void *s2);
-static int prof_self_cmp(const void *s1, const void *s2);
-#endif
 static void funccal_unref(funccall_T *fc, ufunc_T *fp, int force);
 
     void
@@ -1312,6 +1304,12 @@ restore_funccal(void)
     }
 }
 
+    funccall_T *
+get_current_funccal(void)
+{
+    return current_funccal;
+}
+
 #if defined(EXITFREE) || defined(PROTO)
     void
 free_all_functions(void)
@@ -2762,36 +2760,6 @@ get_expanded_name(char_u *name, int check)
 #endif
 
 #if defined(FEAT_PROFILE) || defined(PROTO)
-/*
- * Start profiling function "fp".
- */
-    static void
-func_do_profile(ufunc_T *fp)
-{
-    int		len = fp->uf_lines.ga_len;
-
-    if (!fp->uf_prof_initialized)
-    {
-	if (len == 0)
-	    len = 1;  /* avoid getting error for allocating zero bytes */
-	fp->uf_tm_count = 0;
-	profile_zero(&fp->uf_tm_self);
-	profile_zero(&fp->uf_tm_total);
-	if (fp->uf_tml_count == NULL)
-	    fp->uf_tml_count = ALLOC_CLEAR_MULT(int, len);
-	if (fp->uf_tml_total == NULL)
-	    fp->uf_tml_total = ALLOC_CLEAR_MULT(proftime_T, len);
-	if (fp->uf_tml_self == NULL)
-	    fp->uf_tml_self = ALLOC_CLEAR_MULT(proftime_T, len);
-	fp->uf_tml_idx = -1;
-	if (fp->uf_tml_count == NULL || fp->uf_tml_total == NULL
-						    || fp->uf_tml_self == NULL)
-	    return;	    /* out of memory */
-	fp->uf_prof_initialized = TRUE;
-    }
-
-    fp->uf_profiling = TRUE;
-}
 
 /*
  * Dump the profiling results for all functions in file "fd".
@@ -2871,121 +2839,6 @@ func_dump_profile(FILE *fd)
     vim_free(sorttab);
 }
 
-    static void
-prof_sort_list(
-    FILE	*fd,
-    ufunc_T	**sorttab,
-    int		st_len,
-    char	*title,
-    int		prefer_self)	/* when equal print only self time */
-{
-    int		i;
-    ufunc_T	*fp;
-
-    fprintf(fd, "FUNCTIONS SORTED ON %s TIME\n", title);
-    fprintf(fd, "count  total (s)   self (s)  function\n");
-    for (i = 0; i < 20 && i < st_len; ++i)
-    {
-	fp = sorttab[i];
-	prof_func_line(fd, fp->uf_tm_count, &fp->uf_tm_total, &fp->uf_tm_self,
-								 prefer_self);
-	if (fp->uf_name[0] == K_SPECIAL)
-	    fprintf(fd, " <SNR>%s()\n", fp->uf_name + 3);
-	else
-	    fprintf(fd, " %s()\n", fp->uf_name);
-    }
-    fprintf(fd, "\n");
-}
-
-/*
- * Print the count and times for one function or function line.
- */
-    static void
-prof_func_line(
-    FILE	*fd,
-    int		count,
-    proftime_T	*total,
-    proftime_T	*self,
-    int		prefer_self)	/* when equal print only self time */
-{
-    if (count > 0)
-    {
-	fprintf(fd, "%5d ", count);
-	if (prefer_self && profile_equal(total, self))
-	    fprintf(fd, "           ");
-	else
-	    fprintf(fd, "%s ", profile_msg(total));
-	if (!prefer_self && profile_equal(total, self))
-	    fprintf(fd, "           ");
-	else
-	    fprintf(fd, "%s ", profile_msg(self));
-    }
-    else
-	fprintf(fd, "                            ");
-}
-
-/*
- * Compare function for total time sorting.
- */
-    static int
-prof_total_cmp(const void *s1, const void *s2)
-{
-    ufunc_T	*p1, *p2;
-
-    p1 = *(ufunc_T **)s1;
-    p2 = *(ufunc_T **)s2;
-    return profile_cmp(&p1->uf_tm_total, &p2->uf_tm_total);
-}
-
-/*
- * Compare function for self time sorting.
- */
-    static int
-prof_self_cmp(const void *s1, const void *s2)
-{
-    ufunc_T	*p1, *p2;
-
-    p1 = *(ufunc_T **)s1;
-    p2 = *(ufunc_T **)s2;
-    return profile_cmp(&p1->uf_tm_self, &p2->uf_tm_self);
-}
-
-/*
- * Prepare profiling for entering a child or something else that is not
- * counted for the script/function itself.
- * Should always be called in pair with prof_child_exit().
- */
-    void
-prof_child_enter(
-    proftime_T *tm)	/* place to store waittime */
-{
-    funccall_T *fc = current_funccal;
-
-    if (fc != NULL && fc->func->uf_profiling)
-	profile_start(&fc->prof_child);
-    script_prof_save(tm);
-}
-
-/*
- * Take care of time spent in a child.
- * Should always be called after prof_child_enter().
- */
-    void
-prof_child_exit(
-    proftime_T *tm)	/* where waittime was stored */
-{
-    funccall_T *fc = current_funccal;
-
-    if (fc != NULL && fc->func->uf_profiling)
-    {
-	profile_end(&fc->prof_child);
-	profile_sub_wait(tm, &fc->prof_child); /* don't count waiting time */
-	profile_add(&fc->func->uf_tm_children, &fc->prof_child);
-	profile_add(&fc->func->uf_tml_children, &fc->prof_child);
-    }
-    script_prof_restore(tm);
-}
-
 #endif /* FEAT_PROFILE */
 
 #if defined(FEAT_CMDL_COMPL) || defined(PROTO)
@@ -3574,71 +3427,6 @@ get_func_line(
     return retval;
 }
 
-#if defined(FEAT_PROFILE) || defined(PROTO)
-/*
- * Called when starting to read a function line.
- * "sourcing_lnum" must be correct!
- * When skipping lines it may not actually be executed, but we won't find out
- * until later and we need to store the time now.
- */
-    void
-func_line_start(void *cookie)
-{
-    funccall_T	*fcp = (funccall_T *)cookie;
-    ufunc_T	*fp = fcp->func;
-
-    if (fp->uf_profiling && sourcing_lnum >= 1
-				      && sourcing_lnum <= fp->uf_lines.ga_len)
-    {
-	fp->uf_tml_idx = sourcing_lnum - 1;
-	/* Skip continuation lines. */
-	while (fp->uf_tml_idx > 0 && FUNCLINE(fp, fp->uf_tml_idx) == NULL)
-	    --fp->uf_tml_idx;
-	fp->uf_tml_execed = FALSE;
-	profile_start(&fp->uf_tml_start);
-	profile_zero(&fp->uf_tml_children);
-	profile_get_wait(&fp->uf_tml_wait);
-    }
-}
-
-/*
- * Called when actually executing a function line.
- */
-    void
-func_line_exec(void *cookie)
-{
-    funccall_T	*fcp = (funccall_T *)cookie;
-    ufunc_T	*fp = fcp->func;
-
-    if (fp->uf_profiling && fp->uf_tml_idx >= 0)
-	fp->uf_tml_execed = TRUE;
-}
-
-/*
- * Called when done with a function line.
- */
-    void
-func_line_end(void *cookie)
-{
-    funccall_T	*fcp = (funccall_T *)cookie;
-    ufunc_T	*fp = fcp->func;
-
-    if (fp->uf_profiling && fp->uf_tml_idx >= 0)
-    {
-	if (fp->uf_tml_execed)
-	{
-	    ++fp->uf_tml_count[fp->uf_tml_idx];
-	    profile_end(&fp->uf_tml_start);
-	    profile_sub_wait(&fp->uf_tml_wait, &fp->uf_tml_start);
-	    profile_add(&fp->uf_tml_total[fp->uf_tml_idx], &fp->uf_tml_start);
-	    profile_self(&fp->uf_tml_self[fp->uf_tml_idx], &fp->uf_tml_start,
-							&fp->uf_tml_children);
-	}
-	fp->uf_tml_idx = -1;
-    }
-}
-#endif
-
 /*
  * Return TRUE if the currently active function should be ended, because a
  * return was encountered or an error occurred.  Used inside a ":while".
diff --git a/src/version.c b/src/version.c
index fb7e5c31c..4316068e7 100644
--- a/src/version.c
+++ b/src/version.c
@@ -778,6 +778,8 @@ static char *(features[]) =
 static int included_patches[] =
 {   /* Add new patch number below this line */
 /**/
+    1684,
+/**/
     1683,
 /**/
     1682,