Re: [ast-optimizer-branch]: Simplify STMT_EXPR's

[Daniel Berlin <dberlin at dberlin dot org>]

On Thu, 2 May 2002, Sebastian Pop wrote:

> On Thu, May 02, 2002 at 10:34:44AM -0400, Daniel Berlin wrote:
> > 
> > but during rtl expansion, it aborts failing the following test in 
> > varasm.c:
> > 
> >   /* Check that we are not being given an automatic variable.  */
> >   /* A weak alias has TREE_PUBLIC set but not the other bits.  */
> >   if (TREE_CODE (decl) == PARM_DECL
> >       || TREE_CODE (decl) == RESULT_DECL
> >       || (TREE_CODE (decl) == VAR_DECL
> >           && !TREE_STATIC (decl)
> >           && !TREE_PUBLIC (decl)
> >           && !DECL_EXTERNAL (decl)
> >           && !DECL_REGISTER (decl)))
> >     abort ();
> > 
> That's what happens also when some of the DECL_STMTs were deleted after 
> statement rechaining. 
> 
> Maybe the error comes from the following ? Can you check ?
> 
> TREE_CHAIN (new_stmt) = TREE_CHAIN (temp);
> TREE_CHAIN (temp) = new_stmt;

Doesn't this turn:


temp
stmt_2

into

temp
new_stmt
stmt_2

?

I wrote it around 2am, so it's quite possible i wasn't thinking at all.

Basically, it's trying to turn:


...
x = y;
}


into
...
x = y;
<new temp> = x;
}


Let me see if it works for the case of

...
x;
}

where we transform it into

...
<new temporary> = x;
}

without inserting anything (IE we directly modify the tree for x, 
rather than chain a new stmt in. )



> 
> You cannot really trust on the prety printer since it prints variable declarations
> from the SCOPE_STMT's BLOCK and doesn't print anything for a DECL_STMT.
> So the only way to know wether you removed or not some DECL_STMTs is to use the
> debug_tree ().  Maybe I have to rewrite this portion of the pretty printer 
> since it is missleading.
Ahhhh.
Wasn't aware of that.

> 
> You no longer use 'temp', nor "last" in the following,
> and the code above has no side effects.

YUp, thanks, I wasn't paying attention.

> 
> > + 	  expr_s = NULL_TREE;
> > + 	  add_tree (STMT_EXPR_STMT (expr), pre_p);
> > + 	  return expr_s;
> > + 	}
> So I suggest :
>   else
>     {
>        expr_s = NULL_TREE; 
>        add_tree (STMT_EXPR_STMT (expr), pre_p); 
>        return expr_s; 
>     }
> 
> that can be transformed into :
> else 
>     { 
>        add_tree (STMT_EXPR_STMT (expr), pre_p);
>        return NULL_TREE;
>     }
> >         break;
> >   
> >       case SAVE_EXPR:
> > *************** static int
> > *** 2478,2483 ****
> > --- 2546,2553 ----
> >   expr_has_effect (expr)
> >        tree expr;
> >   {
> > +   if (expr == NULL_TREE)
> > +     return 0;
> >     return (TREE_SIDE_EFFECTS (expr)
> >   	  || (TREE_CODE (expr) == CONVERT_EXPR
> >   	      && VOID_TYPE_P (TREE_TYPE (expr))));
> > 
>