[PATCH] Speed up gimple_tree_eq with release checking

[PATCH] Speed up gimple_tree_eq with release checking

[Richard Guenther]

The gimplifier is still one part of the compiler that is comparatively
slow.  Looking at profiles lookup_tmp_var and friends stick out, the
following makes expensive rehashing for assertion purposes conditional
on checking.

Applied as obvious.

Of course I've always found this poor mans CSE during gimplification
somewhat odd ...

Richard.

2010-12-08  Richard Guenther  <rguenther@suse.de>

	* gimplify.c (gimple_tree_eq): Assert hashes are equal only
	when checking is enabled.

Index: gcc/gimplify.c
===================================================================
*** gcc/gimplify.c	(revision 167583)
--- gcc/gimplify.c	(working copy)
*************** gimple_tree_eq (const void *p1, const vo
*** 145,153 ****
--- 145,155 ----
    if (!operand_equal_p (t1, t2, 0))
      return 0;
  
+ #ifdef ENABLE_CHECKING
    /* Only allow them to compare equal if they also hash equal; otherwise
       results are nondeterminate, and we fail bootstrap comparison.  */
    gcc_assert (gimple_tree_hash (p1) == gimple_tree_hash (p2));
+ #endif
  
    return 1;
  }

Re: [PATCH] Speed up gimple_tree_eq with release checking

[Jakub Jelinek]

On Wed, Dec 08, 2010 at 03:14:13PM +0100, Richard Guenther wrote:
> 
> The gimplifier is still one part of the compiler that is comparatively
> slow.  Looking at profiles lookup_tmp_var and friends stick out, the
> following makes expensive rehashing for assertion purposes conditional
> on checking.
> 
> Applied as obvious.

It should have been gcc_checking_assert instead...

> 2010-12-08  Richard Guenther  <rguenther@suse.de>
> 
> 	* gimplify.c (gimple_tree_eq): Assert hashes are equal only
> 	when checking is enabled.
> 
> Index: gcc/gimplify.c
> ===================================================================
> *** gcc/gimplify.c	(revision 167583)
> --- gcc/gimplify.c	(working copy)
> *************** gimple_tree_eq (const void *p1, const vo
> *** 145,153 ****
> --- 145,155 ----
>     if (!operand_equal_p (t1, t2, 0))
>       return 0;
>   
> + #ifdef ENABLE_CHECKING
>     /* Only allow them to compare equal if they also hash equal; otherwise
>        results are nondeterminate, and we fail bootstrap comparison.  */
>     gcc_assert (gimple_tree_hash (p1) == gimple_tree_hash (p2));
> + #endif
>   
>     return 1;
>   }

	Jakub

Re: [PATCH] Speed up gimple_tree_eq with release checking

[Richard Guenther]

On Wed, 8 Dec 2010, Jakub Jelinek wrote:

> On Wed, Dec 08, 2010 at 03:14:13PM +0100, Richard Guenther wrote:
> > 
> > The gimplifier is still one part of the compiler that is comparatively
> > slow.  Looking at profiles lookup_tmp_var and friends stick out, the
> > following makes expensive rehashing for assertion purposes conditional
> > on checking.
> > 
> > Applied as obvious.
> 
> It should have been gcc_checking_assert instead...

Doesn't that preserve side-effects?

Richard.

> > 2010-12-08  Richard Guenther  <rguenther@suse.de>
> > 
> > 	* gimplify.c (gimple_tree_eq): Assert hashes are equal only
> > 	when checking is enabled.
> > 
> > Index: gcc/gimplify.c
> > ===================================================================
> > *** gcc/gimplify.c	(revision 167583)
> > --- gcc/gimplify.c	(working copy)
> > *************** gimple_tree_eq (const void *p1, const vo
> > *** 145,153 ****
> > --- 145,155 ----
> >     if (!operand_equal_p (t1, t2, 0))
> >       return 0;
> >   
> > + #ifdef ENABLE_CHECKING
> >     /* Only allow them to compare equal if they also hash equal; otherwise
> >        results are nondeterminate, and we fail bootstrap comparison.  */
> >     gcc_assert (gimple_tree_hash (p1) == gimple_tree_hash (p2));
> > + #endif
> >   
> >     return 1;
> >   }
> 
> 	Jakub
> 
> 

-- 
Richard Guenther <rguenther@suse.de>
Novell / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746 - GF: Markus Rex