[PATCH][gdb/symtab] Fix symbol loading performance regression

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

[PATCH][gdb/symtab] Fix symbol loading performance regression

Tom de Vries
Hi,

The commit "[gdb/symtab] Fix language of duplicate static minimal symbol"
introduces a performance regression, when loading a cc1 executable build with
-O0 -g and gcc 7.4.0.  The performance regression, measured in 'real' time is
about 175%.

The slower execution comes from the fact that the fix in symbol_set_names
makes the call to symbol_find_demangled_name unconditional.

Fix this by reverting the commit, and redoing the fix as follows.

Recapturing the original problem, the first time symbol_set_names is called
with gsymbol.language == lang_auto and linkage_name == "_ZL3foov", the name is
not present in the per_bfd->demangled_names_hash hash table, so
symbol_find_demangled_name is called to demangle the name, after which the
mangled/demangled pair is added to the hashtable.  The call to
symbol_find_demangled_name also sets gsymbol.language to lang_cplus.
The second time symbol_set_names is called with gsymbol.language == lang_auto
and linkage_name == "_ZL3foov", the name is present in the hash table, so the
demangled name from the hash table is used.  However, the language of the
symbol remains lang_auto.

Fix this by adding a field language in struct demangled_name_entry, and using
the field in symbol_set_names to set the language of gsymbol, if necessary.

Tested on x86_64-linux.

OK for trunk?

Thanks,
- Tom

[gdb/symtab] Fix symbol loading performance regression

gdb/ChangeLog:

2019-05-11  Tom de Vries  <[hidden email]>

        PR symtab/24545
        * symtab.c (struct demangled_name_entry): Add language field.
        (symbol_set_names):  Revert "[gdb/symtab] Fix language of duplicate
        static minimal symbol".  Set and use language field.

---
 gdb/symtab.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/gdb/symtab.c b/gdb/symtab.c
index 130d5cd48f..44964533ee 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -713,6 +713,7 @@ symbol_set_language (struct general_symbol_info *gsymbol,
 struct demangled_name_entry
 {
   const char *mangled;
+  ENUM_BITFIELD(language) language : LANGUAGE_BITS;
   char demangled[1];
 };
 
@@ -853,11 +854,6 @@ symbol_set_names (struct general_symbol_info *gsymbol,
   else
     linkage_name_copy = linkage_name;
 
-  /* Set the symbol language.  */
-  char *demangled_name_ptr
-    = symbol_find_demangled_name (gsymbol, linkage_name_copy);
-  gdb::unique_xmalloc_ptr<char> demangled_name (demangled_name_ptr);
-
   entry.mangled = linkage_name_copy;
   slot = ((struct demangled_name_entry **)
   htab_find_slot (per_bfd->demangled_names_hash.get (),
@@ -870,7 +866,9 @@ symbol_set_names (struct general_symbol_info *gsymbol,
       || (gsymbol->language == language_go
   && (*slot)->demangled[0] == '\0'))
     {
-      int demangled_len = demangled_name ? strlen (demangled_name.get ()) : 0;
+      char *demangled_name = symbol_find_demangled_name (gsymbol,
+ linkage_name_copy);
+      int demangled_len = demangled_name ? strlen (demangled_name) : 0;
 
       /* Suppose we have demangled_name==NULL, copy_name==0, and
  linkage_name_copy==linkage_name.  In this case, we already have the
@@ -906,12 +904,19 @@ symbol_set_names (struct general_symbol_info *gsymbol,
   strcpy (mangled_ptr, linkage_name_copy);
   (*slot)->mangled = mangled_ptr;
  }
+      (*slot)->language = gsymbol->language;
 
       if (demangled_name != NULL)
- strcpy ((*slot)->demangled, demangled_name.get());
+ {
+  strcpy ((*slot)->demangled, demangled_name);
+  xfree (demangled_name);
+ }
       else
  (*slot)->demangled[0] = '\0';
     }
+  else if (gsymbol->language == language_unknown
+   || gsymbol->language == language_auto)
+    gsymbol->language = (*slot)->language;
 
   gsymbol->name = (*slot)->mangled;
   if ((*slot)->demangled[0] != '\0')
Reply | Threaded
Open this post in threaded view
|

[PING][PATCH][gdb/symtab] Fix symbol loading performance regression

Tom de Vries
On 13-05-19 11:27, Tom de Vries wrote:

> Hi,
>
> The commit "[gdb/symtab] Fix language of duplicate static minimal symbol"
> introduces a performance regression, when loading a cc1 executable build with
> -O0 -g and gcc 7.4.0.  The performance regression, measured in 'real' time is
> about 175%.
>
> The slower execution comes from the fact that the fix in symbol_set_names
> makes the call to symbol_find_demangled_name unconditional.
>
> Fix this by reverting the commit, and redoing the fix as follows.
>
> Recapturing the original problem, the first time symbol_set_names is called
> with gsymbol.language == lang_auto and linkage_name == "_ZL3foov", the name is
> not present in the per_bfd->demangled_names_hash hash table, so
> symbol_find_demangled_name is called to demangle the name, after which the
> mangled/demangled pair is added to the hashtable.  The call to
> symbol_find_demangled_name also sets gsymbol.language to lang_cplus.
> The second time symbol_set_names is called with gsymbol.language == lang_auto
> and linkage_name == "_ZL3foov", the name is present in the hash table, so the
> demangled name from the hash table is used.  However, the language of the
> symbol remains lang_auto.
>
> Fix this by adding a field language in struct demangled_name_entry, and using
> the field in symbol_set_names to set the language of gsymbol, if necessary.
>
> Tested on x86_64-linux.
>
> OK for trunk?
>

Ping.

Thanks,
- Tom

> [gdb/symtab] Fix symbol loading performance regression
>
> gdb/ChangeLog:
>
> 2019-05-11  Tom de Vries  <[hidden email]>
>
> PR symtab/24545
> * symtab.c (struct demangled_name_entry): Add language field.
> (symbol_set_names):  Revert "[gdb/symtab] Fix language of duplicate
> static minimal symbol".  Set and use language field.
>
> ---
>  gdb/symtab.c | 19 ++++++++++++-------
>  1 file changed, 12 insertions(+), 7 deletions(-)
>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index 130d5cd48f..44964533ee 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -713,6 +713,7 @@ symbol_set_language (struct general_symbol_info *gsymbol,
>  struct demangled_name_entry
>  {
>    const char *mangled;
> +  ENUM_BITFIELD(language) language : LANGUAGE_BITS;
>    char demangled[1];
>  };
>  
> @@ -853,11 +854,6 @@ symbol_set_names (struct general_symbol_info *gsymbol,
>    else
>      linkage_name_copy = linkage_name;
>  
> -  /* Set the symbol language.  */
> -  char *demangled_name_ptr
> -    = symbol_find_demangled_name (gsymbol, linkage_name_copy);
> -  gdb::unique_xmalloc_ptr<char> demangled_name (demangled_name_ptr);
> -
>    entry.mangled = linkage_name_copy;
>    slot = ((struct demangled_name_entry **)
>    htab_find_slot (per_bfd->demangled_names_hash.get (),
> @@ -870,7 +866,9 @@ symbol_set_names (struct general_symbol_info *gsymbol,
>        || (gsymbol->language == language_go
>    && (*slot)->demangled[0] == '\0'))
>      {
> -      int demangled_len = demangled_name ? strlen (demangled_name.get ()) : 0;
> +      char *demangled_name = symbol_find_demangled_name (gsymbol,
> + linkage_name_copy);
> +      int demangled_len = demangled_name ? strlen (demangled_name) : 0;
>  
>        /* Suppose we have demangled_name==NULL, copy_name==0, and
>   linkage_name_copy==linkage_name.  In this case, we already have the
> @@ -906,12 +904,19 @@ symbol_set_names (struct general_symbol_info *gsymbol,
>    strcpy (mangled_ptr, linkage_name_copy);
>    (*slot)->mangled = mangled_ptr;
>   }
> +      (*slot)->language = gsymbol->language;
>  
>        if (demangled_name != NULL)
> - strcpy ((*slot)->demangled, demangled_name.get());
> + {
> +  strcpy ((*slot)->demangled, demangled_name);
> +  xfree (demangled_name);
> + }
>        else
>   (*slot)->demangled[0] = '\0';
>      }
> +  else if (gsymbol->language == language_unknown
> +   || gsymbol->language == language_auto)
> +    gsymbol->language = (*slot)->language;
>  
>    gsymbol->name = (*slot)->mangled;
>    if ((*slot)->demangled[0] != '\0')
>
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH][gdb/symtab] Fix symbol loading performance regression

Tom Tromey-2
In reply to this post by Tom de Vries
>>>>> "Tom" == Tom de Vries <[hidden email]> writes:

Tom> 2019-05-11  Tom de Vries  <[hidden email]>

Tom> PR symtab/24545
Tom> * symtab.c (struct demangled_name_entry): Add language field.
Tom> (symbol_set_names):  Revert "[gdb/symtab] Fix language of duplicate
Tom> static minimal symbol".  Set and use language field.

Thanks for doing this.

Tom> +      char *demangled_name = symbol_find_demangled_name (gsymbol,
Tom> + linkage_name_copy);

I think it would be better to do

    gdb::unique_xmalloc_ptr<char> demangled_name (...);

and then adjust the code to use ".get ()" as needed.

This is ok with that change.

Tom
Reply | Threaded
Open this post in threaded view
|

[8.3 backport][gdb/symtab] Fix symbol loading performance regression

Tom de Vries
On 29-05-19 20:36, Tom Tromey wrote:

>>>>>> "Tom" == Tom de Vries <[hidden email]> writes:
>
> Tom> 2019-05-11  Tom de Vries  <[hidden email]>
>
> Tom> PR symtab/24545
> Tom> * symtab.c (struct demangled_name_entry): Add language field.
> Tom> (symbol_set_names):  Revert "[gdb/symtab] Fix language of duplicate
> Tom> static minimal symbol".  Set and use language field.
>
> Thanks for doing this.
>
> Tom> +      char *demangled_name = symbol_find_demangled_name (gsymbol,
> Tom> + linkage_name_copy);
>
> I think it would be better to do
>
>     gdb::unique_xmalloc_ptr<char> demangled_name (...);
>
> and then adjust the code to use ".get ()" as needed.
>
> This is ok with that change.

Is this also ok for 8.3.1? The patch applies cleanly.

Thanks,
- Tom
Reply | Threaded
Open this post in threaded view
|

Re: [8.3 backport][gdb/symtab] Fix symbol loading performance regression

Tom Tromey-2
Tom> Is this also ok for 8.3.1? The patch applies cleanly.

Yes, I think so.  Thanks.

Tom