gprof: only process line numbers for intersection of vmas and histograms

Some programs like RTOS firmware may have a large number of symbols.
The profile information in the profile data file includes histogram
records, which capture low PC and high PC of program execution.  If all
histogram records come in the profile data file before any call-graph
records and basic-block records, we can look up only the line numbers
within low PC and high PC in histogram records, which reduces processing
time for such a firmware from ~2 minutes to ~2 seconds.

Add symbol table access function, get_symtab, get_symtab_direct and
set_symtab to delay loading the symbol table until its first use.

	* aarch64.c (aarch64_find_call): Call get_symtab to get the
	symbol table pointer
	* alpha.c (alpha_find_call): Likewise.
	* basic_blocks.c (bb_read_rec): Likewise.
	(bb_write_blocks): Likewise.
	(print_exec_counts): Likewise.
	(print_annotated_source): Likewise.
	* call_graph.c (cg_tally): Likewise.
	(cg_write_arcs): Likewise.
	* cg_arcs.c (cycle_link): Likewise.
	(propagate_flags): Likewise.
	(cg_assemble): Likewise.
	* cg_print.c (cg_print): Likewise.
	(cg_print_index): Likewise.
	(cg_print_function_ordering): Likewise.
	* corefile.c: Include "gmon_io.h".
	(core_create_syms_from): Call get_symtab_direct to get the
	symbol table pointer.
	(core_create_function_syms): Likewise.
	(core_create_line_syms): Likewise.  If all histogram records
	come in the profile data file before any call-graph records and
	basic-block records, we can look up only the line numbers within
	low PC and high PC in histogram records.
	* gmon_io.c (gmon_histograms_first): New.
	(gmon_out_read): Set gmon_histograms_first to true if all
	histogram records come first.
	(gmon_out_write): Call get_symtab to get the symbol table
	pointer.
	* hist.c (scale_and_align_entries): Likewise.
	(hist_assign_samples_1): Likewise.
	(hist_print): Likewise.
	* i386.c (i386_find_call): Likewise.
	* mips.c (mips_find_call): Likewise.
	* sparc.c (sparc_find_call): Likewise.
	* sym_ids.c (sym_id_parse): Likewise.
	* vax.c (vax_find_call): Likewise.
	* gmon_io.h (gmon_histograms_first): New.
	* gprof.c (man): Don't create profile info.
	(symtab_init): New.
	* gprof.h (symtab_init): New.
	* symtab.c (symtab): Changed to static.
	(get_symtab_direct): New.
	(get_symtab): Likewise.
	(set_symtab): Likewise.
	* symtab.h (symtab): Removed.
	(get_symtab_direct): New.
	(get_symtab): Likewise.
	(set_symtab): Likewise.

Signed-off-by: Richard Allen <rsaxvc@gmail.com>
Co-Authored-By: H.J. Lu <hjl.tools@gmail.com>

gprof/aarch64.c

@@ -50,6 +50,7 @@ aarch64_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   bfd_vma pc, dest_pc, offset;
   unsigned int insn;
   Sym *child;
+  Sym_Table *symtab = get_symtab ();
 
   DBG (CALLDEBUG, printf ("[find_call] %s: 0x%lx to 0x%lx\n",
 			  parent->name, (unsigned long) p_lowpc,
@@ -75,7 +76,7 @@ aarch64_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 
 	  if (hist_check_address (dest_pc))
 	    {
-	      child = sym_lookup (&symtab, dest_pc);
+	      child = sym_lookup (symtab, dest_pc);
 
 	      if (child)
 		{

gprof/alpha.c

@@ -95,6 +95,7 @@ alpha_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   bfd_vma pc, dest_pc;
   unsigned int insn;
   Sym *child;
+  Sym_Table *symtab = get_symtab ();
 
   if (indirect_child.name == NULL)
     {
@@ -149,7 +150,7 @@ alpha_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 			       ^ 0x100000) - 0x100000);
 	  if (hist_check_address (dest_pc))
 	    {
-	      child = sym_lookup (&symtab, dest_pc);
+	      child = sym_lookup (symtab, dest_pc);
               if (child)
                 {
 	          DBG (CALLDEBUG,

gprof/basic_blocks.c

@@ -122,6 +122,7 @@ bb_read_rec (FILE *ifp, const char *filename)
   unsigned int nblocks, b;
   bfd_vma addr, ncalls;
   Sym *sym;
+  Sym_Table *symtab;
 
   if (gmon_io_read_32 (ifp, &nblocks))
     {
@@ -130,6 +131,8 @@ bb_read_rec (FILE *ifp, const char *filename)
       done (1);
     }
 
+  symtab = get_symtab ();
+
   nblocks = bfd_get_32 (core_bfd, (bfd_byte *) & nblocks);
   if (gmon_file_version == 0)
     fskip_string (ifp);
@@ -163,7 +166,7 @@ bb_read_rec (FILE *ifp, const char *filename)
 	 profiling at the line-by-line level:  */
       if (line_granularity)
 	{
-	  sym = sym_lookup (&symtab, addr);
+	  sym = sym_lookup (symtab, addr);
 
 	  if (sym)
 	    {
@@ -210,9 +213,10 @@ bb_write_blocks (FILE *ofp, const char *filename)
   unsigned int nblocks = 0;
   Sym *sym;
   int i;
+  Sym_Table *symtab = get_symtab ();
 
   /* Count how many non-zero blocks with have:  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       for (i = 0; i < NBBS && sym->bb_addr[i]; i++)
 	;
@@ -228,7 +232,7 @@ bb_write_blocks (FILE *ofp, const char *filename)
     }
 
   /* Write counts:  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       for (i = 0; i < NBBS && sym->bb_addr[i]; i++)
 	{
@@ -252,6 +256,7 @@ print_exec_counts (void)
 {
   Sym **sorted_bbs, *sym;
   unsigned int i, j, len;
+  Sym_Table *symtab = get_symtab ();
 
   if (first_output)
     first_output = false;
@@ -259,10 +264,10 @@ print_exec_counts (void)
     printf ("\f\n");
 
   /* Sort basic-blocks according to function name and line number:  */
-  sorted_bbs = (Sym **) xmalloc (symtab.len * sizeof (sorted_bbs[0]));
+  sorted_bbs = (Sym **) xmalloc (symtab->len * sizeof (sorted_bbs[0]));
   len = 0;
 
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       /* Accept symbol if it's in the INCL_EXEC table
 	 or there is no INCL_EXEC table
@@ -461,10 +466,11 @@ print_annotated_source (void)
   Source_File *sf;
   int i, table_len;
   FILE *ofp;
+  Sym_Table *symtab = get_symtab ();
 
   /* Find maximum line number for each source file that user is
      interested in:  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       /* Accept symbol if it's file is known, its line number is
 	 bigger than anything we have seen for that file so far and
@@ -490,7 +496,7 @@ print_annotated_source (void)
     }
 
   /* Count executions per line:  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       if (sym->file && sym->file->num_lines
 	  && (sym_lookup (&syms[INCL_ANNO], sym->addr)

gprof/call_graph.c

@@ -35,9 +35,10 @@ cg_tally (bfd_vma from_pc, bfd_vma self_pc, unsigned long count)
 {
   Sym *parent;
   Sym *child;
+  Sym_Table *symtab = get_symtab ();
 
-  parent = sym_lookup (&symtab, from_pc);
+  parent = sym_lookup (symtab, from_pc);
-  child = sym_lookup (&symtab, self_pc);
+  child = sym_lookup (symtab, self_pc);
 
   if (child == NULL || parent == NULL)
     return;
@@ -51,10 +52,10 @@ cg_tally (bfd_vma from_pc, bfd_vma self_pc, unsigned long count)
 
      For normal profiling, is_func will be set on all symbols, so this
      code will do nothing.  */
-  while (child >= symtab.base && ! child->is_func)
+  while (child >= symtab->base && ! child->is_func)
     --child;
 
-  if (child < symtab.base)
+  if (child < symtab->base)
     return;
 
   /* Keep arc if it is on INCL_ARCS table or if the INCL_ARCS table
@@ -108,8 +109,9 @@ cg_write_arcs (FILE *ofp, const char *filename)
 {
   Arc *arc;
   Sym *sym;
+  Sym_Table *symtab = get_symtab ();
 
-  for (sym = symtab.base; sym < symtab.limit; sym++)
+  for (sym = symtab->base; sym < symtab->limit; sym++)
     {
       for (arc = sym->cg.children; arc; arc = arc->next_child)
 	{

gprof/cg_arcs.c

@@ -275,11 +275,12 @@ cycle_link (void)
   Sym *sym, *cyc, *member;
   Arc *arc;
   int num;
+  Sym_Table *symtab = get_symtab ();
 
   /* count the number of cycles, and initialize the cycle lists: */
 
   num_cycles = 0;
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       /* this is how you find unattached cycles: */
       if (sym->cg.cyc.head == sym && sym->cg.cyc.next)
@@ -300,7 +301,7 @@ cycle_link (void)
    */
   num = 0;
   cyc = cycle_header;
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       if (!(sym->cg.cyc.head == sym && sym->cg.cyc.next != 0))
 	{
@@ -440,9 +441,10 @@ propagate_flags (Sym **symbols)
 {
   int sym_index;
   Sym *old_head, *child;
+  Sym_Table *symtab = get_symtab ();
 
   old_head = 0;
-  for (sym_index = symtab.len - 1; sym_index >= 0; --sym_index)
+  for (sym_index = symtab->len - 1; sym_index >= 0; --sym_index)
     {
       child = symbols[sym_index];
       /*
@@ -597,12 +599,13 @@ cg_assemble (void)
   Sym *parent, **time_sorted_syms, **top_sorted_syms;
   unsigned int sym_index;
   Arc *arc;
+  Sym_Table *symtab = get_symtab ();
 
   /* Initialize various things:
        Zero out child times.
        Count self-recursive calls.
        Indicate that nothing is on cycles.  */
-  for (parent = symtab.base; parent < symtab.limit; parent++)
+  for (parent = symtab->base; parent < symtab->limit; parent++)
     {
       parent->cg.child_time = 0.0;
       arc = arc_lookup (parent, parent);
@@ -633,7 +636,7 @@ cg_assemble (void)
 
   /* Topologically order things.  If any node is unnumbered, number
      it and any of its descendents.  */
-  for (parent = symtab.base; parent < symtab.limit; parent++)
+  for (parent = symtab->base; parent < symtab->limit; parent++)
     {
       if (parent->cg.top_order == DFN_NAN)
 	cg_dfn (parent);
@@ -643,14 +646,14 @@ cg_assemble (void)
   cycle_link ();
 
   /* Sort the symbol table in reverse topological order.  */
-  top_sorted_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  top_sorted_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
-  for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+  for (sym_index = 0; sym_index < symtab->len; ++sym_index)
-    top_sorted_syms[sym_index] = &symtab.base[sym_index];
+    top_sorted_syms[sym_index] = &symtab->base[sym_index];
 
-  qsort (top_sorted_syms, symtab.len, sizeof (Sym *), cmp_topo);
+  qsort (top_sorted_syms, symtab->len, sizeof (Sym *), cmp_topo);
   DBG (DFNDEBUG,
        printf ("[cg_assemble] topological sort listing\n");
-       for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+       for (sym_index = 0; sym_index < symtab->len; ++sym_index)
 	 {
 	   printf ("[cg_assemble] ");
 	   printf ("%d:", top_sorted_syms[sym_index]->cg.top_order);
@@ -668,24 +671,24 @@ cg_assemble (void)
   /* Starting from the topological bottom, propagate children times
      up to parents.  */
   cycle_time ();
-  for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+  for (sym_index = 0; sym_index < symtab->len; ++sym_index)
     propagate_time (top_sorted_syms[sym_index]);
 
   free (top_sorted_syms);
 
   /* Now, sort by CG.PROP.SELF + CG.PROP.CHILD.  Sorting both the regular
      function names and cycle headers.  */
-  time_sorted_syms = (Sym **) xmalloc ((symtab.len + num_cycles) * sizeof (Sym *));
+  time_sorted_syms = (Sym **) xmalloc ((symtab->len + num_cycles) * sizeof (Sym *));
-  for (sym_index = 0; sym_index < symtab.len; sym_index++)
+  for (sym_index = 0; sym_index < symtab->len; sym_index++)
-    time_sorted_syms[sym_index] = &symtab.base[sym_index];
+    time_sorted_syms[sym_index] = &symtab->base[sym_index];
 
   for (sym_index = 1; sym_index <= num_cycles; sym_index++)
-    time_sorted_syms[symtab.len + sym_index - 1] = &cycle_header[sym_index];
+    time_sorted_syms[symtab->len + sym_index - 1] = &cycle_header[sym_index];
 
-  qsort (time_sorted_syms, symtab.len + num_cycles, sizeof (Sym *),
+  qsort (time_sorted_syms, symtab->len + num_cycles, sizeof (Sym *),
 	 cmp_total);
 
-  for (sym_index = 0; sym_index < symtab.len + num_cycles; sym_index++)
+  for (sym_index = 0; sym_index < symtab->len + num_cycles; sym_index++)
     time_sorted_syms[sym_index]->cg.index = sym_index + 1;
 
   return time_sorted_syms;

gprof/cg_print.c

@@ -504,13 +504,14 @@ cg_print (Sym ** timesortsym)
 {
   unsigned int sym_index;
   Sym *parent;
+  Sym_Table *symtab = get_symtab ();
 
   if (print_descriptions && bsd_style_output)
     bsd_callg_blurb (stdout);
 
   print_header ();
 
-  for (sym_index = 0; sym_index < symtab.len + num_cycles; ++sym_index)
+  for (sym_index = 0; sym_index < symtab->len + num_cycles; ++sym_index)
     {
       parent = timesortsym[sym_index];
 
@@ -570,18 +571,19 @@ cg_print_index (void)
   const char *filename;
   char buf[20];
   int column_width = (output_width - 1) / 3;	/* Don't write in last col!  */
+  Sym_Table *symtab = get_symtab ();
 
   /* Now, sort regular function name
      alphabetically to create an index.  */
-  name_sorted_syms = (Sym **) xmalloc ((symtab.len + num_cycles) * sizeof (Sym *));
+  name_sorted_syms = (Sym **) xmalloc ((symtab->len + num_cycles) * sizeof (Sym *));
 
-  for (sym_index = 0, nnames = 0; sym_index < symtab.len; sym_index++)
+  for (sym_index = 0, nnames = 0; sym_index < symtab->len; sym_index++)
     {
-      if (ignore_zeros && symtab.base[sym_index].ncalls == 0
+      if (ignore_zeros && symtab->base[sym_index].ncalls == 0
-	  && symtab.base[sym_index].hist.time == 0)
+	  && symtab->base[sym_index].hist.time == 0)
 	continue;
 
-      name_sorted_syms[nnames++] = &symtab.base[sym_index];
+      name_sorted_syms[nnames++] = &symtab->base[sym_index];
     }
 
   qsort (name_sorted_syms, nnames, sizeof (Sym *), cmp_name);
@@ -787,6 +789,7 @@ cg_print_function_ordering (void)
 #endif
   Sym **unused_syms, **used_syms, **scratch_syms;
   Arc **unplaced_arcs, **high_arcs, **scratch_arcs;
+  Sym_Table *symtab = get_symtab ();
 
   sym_index = 0;
   used = 0;
@@ -797,29 +800,29 @@ cg_print_function_ordering (void)
   scratch_arc_count = 0;
 
   /* First group all the unused functions together.  */
-  unused_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  unused_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
-  used_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  used_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
-  scratch_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  scratch_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
   high_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
   scratch_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
   unplaced_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
 
   /* Walk through all the functions; mark those which are never
      called as placed (we'll emit them as a group later).  */
-  for (sym_index = 0, used = 0, unused = 0; sym_index < symtab.len; sym_index++)
+  for (sym_index = 0, used = 0, unused = 0; sym_index < symtab->len; sym_index++)
     {
-      if (symtab.base[sym_index].ncalls == 0)
+      if (symtab->base[sym_index].ncalls == 0)
 	{
-	  unused_syms[unused++] = &symtab.base[sym_index];
+	  unused_syms[unused++] = &symtab->base[sym_index];
-	  symtab.base[sym_index].has_been_placed = 1;
+	  symtab->base[sym_index].has_been_placed = 1;
 	}
       else
 	{
-	  used_syms[used++] = &symtab.base[sym_index];
+	  used_syms[used++] = &symtab->base[sym_index];
-	  symtab.base[sym_index].has_been_placed = 0;
+	  symtab->base[sym_index].has_been_placed = 0;
-	  symtab.base[sym_index].next = 0;
+	  symtab->base[sym_index].next = 0;
-	  symtab.base[sym_index].prev = 0;
+	  symtab->base[sym_index].prev = 0;
-	  symtab.base[sym_index].nuses = 0;
+	  symtab->base[sym_index].nuses = 0;
 	}
     }
 
@@ -961,9 +964,9 @@ cg_print_function_ordering (void)
   for (sym_index = 0; sym_index < unused; sym_index++)
     printf("%s\n", unused_syms[sym_index]->name);
 
-  unused_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  unused_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
-  used_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  used_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
-  scratch_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  scratch_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
   high_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
   scratch_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
   unplaced_arcs = (Arc **) xmalloc (numarcs * sizeof (Arc *));
@@ -1236,6 +1239,7 @@ cg_print_file_ordering (void)
   unsigned long sym_index;
   Arc **scratch_arcs;
   char *last;
+  Sym_Table *symtab = get_symtab ();
 
   scratch_arc_count = 0;
 
@@ -1251,11 +1255,11 @@ cg_print_file_ordering (void)
 				    scratch_arcs, &scratch_arc_count);
 
   /* Output .o's not handled by the main placement algorithm.  */
-  for (sym_index = 0; sym_index < symtab.len; sym_index++)
+  for (sym_index = 0; sym_index < symtab->len; sym_index++)
     {
-      if (symtab.base[sym_index].mapped
+      if (symtab->base[sym_index].mapped
-	  && ! symtab.base[sym_index].has_been_placed)
+	  && ! symtab->base[sym_index].has_been_placed)
-	printf ("%s\n", symtab.base[sym_index].name);
+	printf ("%s\n", symtab->base[sym_index].name);
     }
 
   qsort (symbol_map, symbol_map_count, sizeof (struct function_map), cmp_symbol_map);
@@ -1271,19 +1275,19 @@ cg_print_file_ordering (void)
       if (last && !filename_cmp (last, symbol_map[sym_index].file_name))
 	continue;
 
-      for (index2 = 0; index2 < symtab.len; index2++)
+      for (index2 = 0; index2 < symtab->len; index2++)
 	{
-	  if (! symtab.base[index2].mapped)
+	  if (! symtab->base[index2].mapped)
 	    continue;
 
-	  if (!filename_cmp (symtab.base[index2].name,
+	  if (!filename_cmp (symtab->base[index2].name,
 			     symbol_map[sym_index].file_name))
 	    break;
 	}
 
       /* If we didn't find it in the symbol table, then it must
 	 be a .o with no text symbols.  Output it last.  */
-      if (index2 == symtab.len)
+      if (index2 == symtab->len)
 	printf ("%s\n", symbol_map[sym_index].file_name);
       last = symbol_map[sym_index].file_name;
     }

gprof/corefile.c

@@ -27,6 +27,7 @@
 #include "symtab.h"
 #include "hist.h"
 #include "corefile.h"
+#include "gmon_io.h"
 #include "safe-ctype.h"
 #include <limits.h>    /* For UINT_MAX.  */
 
@@ -523,6 +524,7 @@ core_create_syms_from (const char * sym_table_file)
 {
   char type;
   FILE * f;
+  Sym_Table *symtab;
 
   f = fopen (sym_table_file, "r");
   if (!f)
@@ -531,25 +533,27 @@ core_create_syms_from (const char * sym_table_file)
       done (1);
     }
 
-  /* Pass 1 - determine upper bound on number of function names.  */
+  symtab = get_symtab_direct ();
-  symtab.len = num_of_syms_in (f);
 
-  if (symtab.len == 0)
+  /* Pass 1 - determine upper bound on number of function names.  */
+  symtab->len = num_of_syms_in (f);
+
+  if (symtab->len == 0)
     {
       fprintf (stderr, _("%s: file `%s' has no symbols\n"), whoami, sym_table_file);
       done (1);
     }
-  else if (symtab.len == -1U)
+  else if (symtab->len == -1U)
     {
       fprintf (stderr, _("%s: file `%s' has too many symbols\n"),
 	       whoami, sym_table_file);
       done (1);
     }
 
-  symtab.base = (Sym *) xmalloc (symtab.len * sizeof (Sym));
+  symtab->base = (Sym *) xmalloc (symtab->len * sizeof (Sym));
 
   /* Pass 2 - create symbols.  */
-  symtab.limit = symtab.base;
+  symtab->limit = symtab->base;
 
   if (fseek (f, 0, SEEK_SET) != 0)
     {
@@ -564,25 +568,25 @@ core_create_syms_from (const char * sym_table_file)
       if (type != 't' && type != 'T')
 	continue;
 
-      sym_init (symtab.limit);
+      sym_init (symtab->limit);
 
       uint64_t addr;
       sscanf (address, "%" SCNx64, &addr);
-      symtab.limit->addr = addr;
+      symtab->limit->addr = addr;
 
-      symtab.limit->name = (char *) xmalloc (strlen (name) + 1);
+      symtab->limit->name = (char *) xmalloc (strlen (name) + 1);
-      strcpy ((char *) symtab.limit->name, name);
+      strcpy ((char *) symtab->limit->name, name);
-      symtab.limit->mapped = 0;
+      symtab->limit->mapped = 0;
-      symtab.limit->is_func = true;
+      symtab->limit->is_func = true;
-      symtab.limit->is_bb_head = true;
+      symtab->limit->is_bb_head = true;
-      symtab.limit->is_static = (type == 't');
+      symtab->limit->is_static = (type == 't');
 
-      ++symtab.limit;
+      ++symtab->limit;
     }
   fclose (f);
 
-  symtab.len = symtab.limit - symtab.base;
+  symtab->len = symtab->limit - symtab->base;
-  symtab_finalize (&symtab);
+  symtab_finalize (symtab);
 }
 
 static int
@@ -601,6 +605,7 @@ core_create_function_syms (void)
   long i;
   struct function_map * found = NULL;
   int core_has_func_syms = 0;
+  Sym_Table *symtab = get_symtab_direct ();
 
   switch (core_bfd->xvec->flavour)
     {
@@ -615,7 +620,7 @@ core_create_function_syms (void)
     }
 
   /* Pass 1 - determine upper bound on number of function names.  */
-  symtab.len = 0;
+  symtab->len = 0;
 
   for (i = 0; i < core_num_syms; ++i)
     {
@@ -634,19 +639,19 @@ core_create_function_syms (void)
 	     sizeof (struct function_map), search_mapped_symbol);
 	}
       if (found == NULL || found->is_first)
-	++symtab.len;
+	++symtab->len;
     }
 
-  if (symtab.len == 0)
+  if (symtab->len == 0)
     {
       fprintf (stderr, _("%s: file `%s' has no symbols\n"), whoami, a_out_name);
       done (1);
     }
 
-  symtab.base = (Sym *) xmalloc (symtab.len * sizeof (Sym));
+  symtab->base = (Sym *) xmalloc (symtab->len * sizeof (Sym));
 
   /* Pass 2 - create symbols.  */
-  symtab.limit = symtab.base;
+  symtab->limit = symtab->base;
 
   for (i = 0; i < core_num_syms; ++i)
     {
@@ -674,23 +679,23 @@ core_create_function_syms (void)
       if (found && ! found->is_first)
 	continue;
 
-      sym_init (symtab.limit);
+      sym_init (symtab->limit);
 
       /* Symbol offsets are always section-relative.  */
       sym_sec = core_syms[i]->section;
-      symtab.limit->addr = core_syms[i]->value;
+      symtab->limit->addr = core_syms[i]->value;
       if (sym_sec)
-	symtab.limit->addr += bfd_section_vma (sym_sec);
+	symtab->limit->addr += bfd_section_vma (sym_sec);
 
       if (found)
 	{
-	  symtab.limit->name = found->file_name;
+	  symtab->limit->name = found->file_name;
-	  symtab.limit->mapped = 1;
+	  symtab->limit->mapped = 1;
 	}
       else
 	{
-	  symtab.limit->name = core_syms[i]->name;
+	  symtab->limit->name = core_syms[i]->name;
-	  symtab.limit->mapped = 0;
+	  symtab->limit->mapped = 0;
 	}
 
       /* Lookup filename and line number, if we can.  */
@@ -698,10 +703,10 @@ core_create_function_syms (void)
 	const char * filename;
 	const char * func_name;
 
-	if (get_src_info (symtab.limit->addr, & filename, & func_name,
+	if (get_src_info (symtab->limit->addr, & filename, & func_name,
-			  & symtab.limit->line_num))
+			  & symtab->limit->line_num))
 	  {
-	    symtab.limit->file = source_file_lookup_path (filename);
+	    symtab->limit->file = source_file_lookup_path (filename);
 
 	    /* FIXME: Checking __osf__ here does not work with a cross
 	       gprof.  */
@@ -713,36 +718,36 @@ core_create_function_syms (void)
 	       labels do not appear in the symbol table info, so this isn't
 	       necessary.  */
 
-	    if (strcmp (symtab.limit->name, func_name) != 0)
+	    if (strcmp (symtab->limit->name, func_name) != 0)
 	      {
 		/* The symbol's address maps to a different name, so
 		   it can't be a function-entry point.  This happens
 		   for labels, for example.  */
 		DBG (AOUTDEBUG,
 		     printf ("[core_create_function_syms: rej %s (maps to %s)\n",
-			     symtab.limit->name, func_name));
+			     symtab->limit->name, func_name));
 		continue;
 	      }
 #endif
 	  }
       }
 
-      symtab.limit->is_func = (!core_has_func_syms
+      symtab->limit->is_func = (!core_has_func_syms
 			       || (core_syms[i]->flags & BSF_FUNCTION) != 0);
-      symtab.limit->is_bb_head = true;
+      symtab->limit->is_bb_head = true;
 
       if (cxxclass == 't')
-	symtab.limit->is_static = true;
+	symtab->limit->is_static = true;
 
       DBG (AOUTDEBUG, printf ("[core_create_function_syms] %ld %s 0x%lx\n",
-			      (long) (symtab.limit - symtab.base),
+			      (long) (symtab->limit - symtab->base),
-			      symtab.limit->name,
+			      symtab->limit->name,
-			      (unsigned long) symtab.limit->addr));
+			      (unsigned long) symtab->limit->addr));
-      ++symtab.limit;
+      ++symtab->limit;
     }
 
-  symtab.len = symtab.limit - symtab.base;
+  symtab->len = symtab->limit - symtab->base;
-  symtab_finalize (&symtab);
+  symtab_finalize (symtab);
 }
 
 /* Read in symbol table from core.
@@ -755,8 +760,10 @@ core_create_line_syms (void)
   Sym prev, *sym;
   const char *filename;
   Sym_Table ltab;
-  bfd_vma vma_high;
   size_t ltab_reserved;
+  Sym_Table *symtab = get_symtab_direct ();
+  bfd_vma bfd_vma_low = core_text_sect->vma;
+  bfd_vma bfd_vma_high = bfd_vma_low + bfd_section_size (core_text_sect);
 
   /* Create symbols for functions as usual.  This is necessary in
      cases where parts of a program were not compiled with -g.  For
@@ -786,19 +793,82 @@ core_create_line_syms (void)
      lot cleaner now.  */
   memset (&prev, 0, sizeof (prev));
 
-  vma_high = core_text_sect->vma + bfd_section_size (core_text_sect);
+  /* The profile information in the profile data file includes histogram
-  for (vma = core_text_sect->vma; vma < vma_high; vma += insn_boundary)
+     records, which capture low PC and high PC of program execution.  If
+     all histogram records come in the profile data file before any
+     call-graph records and basic-block records, we can look up only the
+     line numbers within low PC and high PC in histogram records.  */
+  if (gmon_histograms_first)
+    for (size_t i = 0; i < num_histograms; ++i)
+      {
+	bfd_vma hist_vma_high = histograms[i].highpc;
+	bfd_vma vma_low = MAX (histograms[i].lowpc, bfd_vma_low);
+	bfd_vma vma_high = MIN (bfd_vma_high, hist_vma_high);
+	for (vma = vma_low; vma < vma_high; vma += insn_boundary)
 	  {
 	    if (ltab.len >= ltab_reserved)
 	      {
 		/* Reserve more space for line symbols.  */
 		ltab_reserved *= 2;
-	  ltab.base = (Sym *) xrealloc (ltab.base, ltab_reserved * sizeof (Sym));
+		ltab.base = xrealloc (ltab.base,
+				      ltab_reserved * sizeof (Sym));
 		ltab.limit = ltab.base + ltab.len;
 	      }
 	    sym_init (ltab.limit);
 
-      if (!get_src_info (vma, &filename, &ltab.limit->name, &ltab.limit->line_num)
+	    if (!get_src_info (vma, &filename, &ltab.limit->name,
+			       &ltab.limit->line_num)
+		|| (prev.name && prev.line_num == ltab.limit->line_num
+		    && strcmp (prev.name, ltab.limit->name) == 0
+		    && filename_cmp (prev.file->name, filename) == 0))
+	      continue;
+
+	    /* Make name pointer a malloc'ed string.  */
+	    ltab.limit->name = xstrdup (ltab.limit->name);
+	    ltab.limit->file = source_file_lookup_path (filename);
+
+	    ltab.limit->addr = vma;
+
+	    /* Set is_static based on the enclosing function, using either:
+	       1) the previous symbol, if it's from the same function, or
+	       2) a symtab lookup.  */
+	    if (prev.name && ltab.limit->file == prev.file
+		&& strcmp (ltab.limit->name, prev.name) == 0)
+	      {
+		ltab.limit->is_static = prev.is_static;
+	      }
+	    else
+	      {
+		sym = sym_lookup (symtab, ltab.limit->addr);
+		if (sym)
+		  ltab.limit->is_static = sym->is_static;
+	      }
+
+	    prev = *ltab.limit;
+
+	    DBG (AOUTDEBUG, printf ("[core_create_line_syms] %lu %s 0x%lx\n",
+				    (unsigned long) (ltab.limit - ltab.base),
+				    ltab.limit->name,
+				    (unsigned long) ltab.limit->addr));
+	    ++ltab.limit;
+	    ++ltab.len;
+	  }
+      }
+  else
+    for (vma = bfd_vma_low; vma < bfd_vma_high; vma += insn_boundary)
+      {
+	if (ltab.len >= ltab_reserved)
+	  {
+	    /* Reserve more space for line symbols.  */
+	    ltab_reserved *= 2;
+	    ltab.base = (Sym *) xrealloc (ltab.base,
+					  ltab_reserved * sizeof (Sym));
+	    ltab.limit = ltab.base + ltab.len;
+	  }
+	sym_init (ltab.limit);
+
+	if (!get_src_info (vma, &filename, &ltab.limit->name,
+			   &ltab.limit->line_num)
 	    || (prev.name && prev.line_num == ltab.limit->line_num
 		&& strcmp (prev.name, ltab.limit->name) == 0
 		&& filename_cmp (prev.file->name, filename) == 0))
@@ -820,7 +890,7 @@ core_create_line_syms (void)
 	  }
 	else
 	  {
-	  sym = sym_lookup(&symtab, ltab.limit->addr);
+	    sym = sym_lookup (symtab, ltab.limit->addr);
 	    if (sym)
 	      ltab.limit->is_static = sym->is_static;
 	  }
@@ -836,17 +906,17 @@ core_create_line_syms (void)
       }
 
   /* Reserve space for function symbols and/or trim excess space.  */
-  ltab_reserved = ltab.len + symtab.len;
+  ltab_reserved = ltab.len + symtab->len;
   ltab.base = xrealloc (ltab.base, ltab_reserved * sizeof (Sym));
   ltab.limit = ltab.base + ltab.len;
 
   /* Copy in function symbols.  */
-  memcpy (ltab.limit, symtab.base, symtab.len * sizeof (Sym));
+  memcpy (ltab.limit, symtab->base, symtab->len * sizeof (Sym));
-  ltab.limit += symtab.len;
+  ltab.limit += symtab->len;
-  ltab.len += symtab.len;
+  ltab.len += symtab->len;
 
   /* Finalize ltab and make it symbol table.  */
   symtab_finalize (&ltab);
-  free (symtab.base);
+  free (symtab->base);
-  symtab = ltab;
+  set_symtab (&ltab);
 }

gprof/gmon_io.c

@@ -58,6 +58,10 @@ static int gmon_write_raw_arc
 int gmon_input = 0;
 int gmon_file_version = 0;	/* 0 == old (non-versioned) file format.  */
 
+/* True if all histogram records come before any call-graph records and
+   basic-block records.  */
+bool gmon_histograms_first = false;
+
 static enum gmon_ptr_size
 gmon_get_ptr_size (void)
 {
@@ -327,6 +331,10 @@ gmon_out_read (const char *filename)
 	  switch (tag)
 	    {
 	    case GMON_TAG_TIME_HIST:
+	      if (narcs || nbbs)
+		gmon_histograms_first = false;
+	      else
+		gmon_histograms_first = true;
 	      ++nhist;
 	      gmon_input |= INPUT_HISTOGRAM;
 	      hist_read_rec (ifp, filename);
@@ -576,6 +584,7 @@ gmon_out_write (const char *filename)
 {
   FILE *ofp;
   struct gmon_hdr ghdr;
+  Sym_Table *symtab;
 
   ofp = fopen (filename, FOPEN_WB);
   if (!ofp)
@@ -584,6 +593,8 @@ gmon_out_write (const char *filename)
       done (1);
     }
 
+  symtab = get_symtab ();
+
   if (file_format == FF_AUTO || file_format == FF_MAGIC)
     {
       /* Write gmon header.  */
@@ -704,7 +715,7 @@ gmon_out_write (const char *filename)
 	}
 
       /* Dump the normalized raw arc information.  */
-      for (sym = symtab.base; sym < symtab.limit; ++sym)
+      for (sym = symtab->base; sym < symtab->limit; ++sym)
 	{
 	  for (arc = sym->cg.children; arc; arc = arc->next_child)
 	    {

gprof/gmon_io.h

@@ -28,6 +28,7 @@
 
 extern int gmon_input;		/* What input did we see?  */
 extern int gmon_file_version;	/* File version are we dealing with.  */
+extern bool gmon_histograms_first;
 
 extern int gmon_io_read_vma (FILE *ifp, bfd_vma *valp);
 extern int gmon_io_read_32 (FILE *ifp, unsigned int *valp);

gprof/gprof.c

@@ -527,17 +527,6 @@ This program is free software.  This program has absolutely no warranty.\n"));
   if (ignore_direct_calls)
     core_get_text_space (core_bfd);
 
-  /* Create symbols from core image.  */
-  if (external_symbol_table)
-    core_create_syms_from (external_symbol_table);
-  else if (line_granularity)
-    core_create_line_syms ();
-  else
-    core_create_function_syms ();
-
-  /* Translate sym specs into syms.  */
-  sym_id_parse ();
-
   if (file_format == FF_PROF)
     {
       fprintf (stderr,
@@ -644,6 +633,23 @@ This program is free software.  This program has absolutely no warranty.\n"));
   return 0;
 }
 
+/* Initialize the symbol table.  */
+
+void
+symtab_init (void)
+{
+  /* Create symbols from core image.  */
+  if (external_symbol_table)
+    core_create_syms_from (external_symbol_table);
+  else if (line_granularity)
+    core_create_line_syms ();
+  else
+    core_create_function_syms ();
+
+  /* Translate sym specs into syms.  */
+  sym_id_parse ();
+}
+
 void
 done (int status)
 {

gprof/gprof.h

@@ -132,4 +132,6 @@ extern bool first_output;		/* no output so far? */
 
 extern void done (int status) ATTRIBUTE_NORETURN;
 
+extern void symtab_init (void);
+
 #endif /* gprof_h */

gprof/hist.c

@@ -294,8 +294,9 @@ scale_and_align_entries (void)
   Sym *sym;
   bfd_vma bin_of_entry;
   bfd_vma bin_of_code;
+  Sym_Table *symtab = get_symtab ();
 
-  for (sym = symtab.base; sym < symtab.limit; sym++)
+  for (sym = symtab->base; sym < symtab->limit; sym++)
     {
       histogram *r = find_histogram_for_pc (sym->addr);
 
@@ -366,6 +367,7 @@ hist_assign_samples_1 (histogram *r)
   unsigned int bin_count;
   unsigned int i, j, k;
   double count_time, credit;
+  Sym_Table *symtab = get_symtab ();
 
   bfd_vma lowpc = r->lowpc / sizeof (UNIT);
 
@@ -392,10 +394,10 @@ hist_assign_samples_1 (histogram *r)
 
          PR gprof/13325: Make sure that K does not get decremented
 	 and J will never be less than 0.  */
-      for (j = k - 1; j < symtab.len; k = ++j)
+      for (j = k - 1; j < symtab->len; k = ++j)
 	{
-	  sym_low_pc = symtab.base[j].hist.scaled_addr;
+	  sym_low_pc = symtab->base[j].hist.scaled_addr;
-	  sym_high_pc = symtab.base[j + 1].hist.scaled_addr;
+	  sym_high_pc = symtab->base[j + 1].hist.scaled_addr;
 
 	  /* If high end of bin is below entry address,
 	     go for next bin.  */
@@ -414,12 +416,12 @@ hist_assign_samples_1 (histogram *r)
 	      DBG (SAMPLEDEBUG,
 		   printf (
 	       "[assign_samples] [0x%lx,0x%lx) %s gets %f ticks %ld overlap\n",
-			   (unsigned long) symtab.base[j].addr,
+			   (unsigned long) symtab->base[j].addr,
 			   (unsigned long) (sizeof (UNIT) * sym_high_pc),
-			   symtab.base[j].name, overlap * count_time / hist_scale,
+			   symtab->base[j].name, overlap * count_time / hist_scale,
 			   (long) overlap));
 
-	      addr = symtab.base[j].addr;
+	      addr = symtab->base[j].addr;
 	      credit = overlap * count_time / hist_scale;
 
 	      /* Credit symbol if it appears in INCL_FLAT or that
@@ -429,7 +431,7 @@ hist_assign_samples_1 (histogram *r)
 		  || (syms[INCL_FLAT].len == 0
 		      && !sym_lookup (&syms[EXCL_FLAT], addr)))
 		{
-		  symtab.base[j].hist.time += credit;
+		  symtab->base[j].hist.time += credit;
 		}
 	      else
 		{
@@ -569,6 +571,7 @@ hist_print (void)
   unsigned log_scale;
   double top_time;
   bfd_vma addr;
+  Sym_Table *symtab = get_symtab ();
 
   if (first_output)
     first_output = false;
@@ -592,12 +595,12 @@ hist_print (void)
 
   /* Sort the symbol table by time (call-count and name as secondary
      and tertiary keys).  */
-  time_sorted_syms = (Sym **) xmalloc (symtab.len * sizeof (Sym *));
+  time_sorted_syms = (Sym **) xmalloc (symtab->len * sizeof (Sym *));
 
-  for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+  for (sym_index = 0; sym_index < symtab->len; ++sym_index)
-    time_sorted_syms[sym_index] = &symtab.base[sym_index];
+    time_sorted_syms[sym_index] = &symtab->base[sym_index];
 
-  qsort (time_sorted_syms, symtab.len, sizeof (Sym *), cmp_time);
+  qsort (time_sorted_syms, symtab->len, sizeof (Sym *), cmp_time);
 
   if (bsd_style_output)
     {
@@ -611,7 +614,7 @@ hist_print (void)
       top_dog = 0;
       top_time = 0.0;
 
-      for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+      for (sym_index = 0; sym_index < symtab->len; ++sym_index)
 	{
 	  sym = time_sorted_syms[sym_index];
 
@@ -648,7 +651,7 @@ hist_print (void)
      I-cache misses etc.).  */
   print_header (SItab[log_scale].prefix);
 
-  for (sym_index = 0; sym_index < symtab.len; ++sym_index)
+  for (sym_index = 0; sym_index < symtab->len; ++sym_index)
     {
       addr = time_sorted_syms[sym_index]->addr;
 

gprof/i386.c

@@ -52,6 +52,7 @@ i386_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   unsigned char *instructp;
   Sym *child;
   bfd_vma pc, destpc;
+  Sym_Table *symtab = get_symtab ();
 
   DBG (CALLDEBUG, printf ("[findcall] %s: 0x%lx to 0x%lx\n",
 			  parent->name, (unsigned long) p_lowpc,
@@ -76,7 +77,7 @@ i386_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 	  destpc = bfd_get_32 (core_bfd, instructp + 1) + pc + 5;
 	  if (hist_check_address (destpc))
 	    {
-	      child = sym_lookup (&symtab, destpc);
+	      child = sym_lookup (symtab, destpc);
 	      if (child && child->addr == destpc)
 		{
 		  /*

gprof/mips.c

@@ -46,6 +46,7 @@ mips_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   int offset;
   Sym *child;
   static bool inited = false;
+  Sym_Table *symtab = get_symtab ();
 
   if (!inited)
     {
@@ -75,7 +76,7 @@ mips_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 	  dest_pc = (pc & ~(bfd_vma) 0xfffffff) | offset;
 	  if (hist_check_address (dest_pc))
 	    {
-	      child = sym_lookup (&symtab, dest_pc);
+	      child = sym_lookup (symtab, dest_pc);
               if (child)
 		{
 	          DBG (CALLDEBUG,

gprof/sparc.c

@@ -47,6 +47,7 @@ sparc_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   bfd_vma pc, dest_pc;
   unsigned int insn;
   Sym *child;
+  Sym_Table *symtab = get_symtab ();
 
   DBG (CALLDEBUG, printf ("[find_call] %s: 0x%lx to 0x%lx\n",
 			  parent->name, (unsigned long) p_lowpc,
@@ -69,7 +70,7 @@ sparc_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 			   ^ 0x20000000) - 0x20000000);
 	  if (hist_check_address (dest_pc))
 	    {
-	      child = sym_lookup (&symtab, dest_pc);
+	      child = sym_lookup (symtab, dest_pc);
 	      if (child)
 		{
 	          DBG (CALLDEBUG,

gprof/sym_ids.c

@@ -278,13 +278,14 @@ sym_id_parse (void)
   Sym *sym, *left, *right;
   struct sym_id *id;
   Sym_Table *tab;
+  Sym_Table *symtab = get_symtab_direct ();
 
   /* Convert symbol ids into Syms, so we can deal with them more easily.  */
   for (id = id_list; id; id = id->next)
     parse_id (id);
 
   /* First determine size of each table.  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       for (id = id_list; id; id = id->next)
 	{
@@ -315,7 +316,7 @@ sym_id_parse (void)
     }
 
   /* Make a second pass through symtab, creating syms as necessary.  */
-  for (sym = symtab.base; sym < symtab.limit; ++sym)
+  for (sym = symtab->base; sym < symtab->limit; ++sym)
     {
       for (id = id_list; id; id = id->next)
 	{

gprof/symtab.c

@@ -28,8 +28,40 @@
 
 static int cmp_addr (const void *, const void *);
 
-Sym_Table symtab;
+/* The symbol table.  */
+static Sym_Table symtab;
 
+/* Return the pointer to the symbol table.  */
+
+Sym_Table *
+get_symtab_direct (void)
+{
+  return &symtab;
+}
+
+/* Return the pointer to the symbol table and initialize it if it isn't
+   initialized yet.  */
+
+Sym_Table *
+get_symtab (void)
+{
+  static Sym_Table *symtab_p;
+  if (!symtab_p)
+    {
+      symtab_init ();
+
+      symtab_p = &symtab;
+    }
+  return symtab_p;
+}
+
+/* Set the symbol table to *LTAB.  */
+
+void
+set_symtab (Sym_Table *ltab)
+{
+  symtab = *ltab;
+}
 
 /* Initialize a symbol (so it's empty).  */
 

gprof/symtab.h

@@ -110,7 +110,9 @@ typedef struct
   }
 Sym_Table;
 
-extern Sym_Table symtab;	/* The symbol table.  */
+extern Sym_Table *get_symtab (void);
+extern Sym_Table *get_symtab_direct (void);
+extern void set_symtab       (Sym_Table *);
 
 extern void sym_init        (Sym *);
 extern void symtab_finalize (Sym_Table *);

gprof/vax.c

@@ -238,6 +238,7 @@ vax_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
   operandenum firstmode;
   bfd_vma pc, destpc;
   static bool inited = false;
+  Sym_Table *symtab = get_symtab ();
 
   if (!inited)
     {
@@ -321,7 +322,7 @@ vax_find_call (Sym *parent, bfd_vma p_lowpc, bfd_vma p_highpc)
 	      destpc = pc + vax_offset (operand);
 	      if (hist_check_address (destpc))
 		{
-		  child = sym_lookup (&symtab, destpc);
+		  child = sym_lookup (symtab, destpc);
 		  if (child)
 		    {
 		      DBG (CALLDEBUG,