Non-call exceptions versus cse

[Andrew Haley <aph at redhat dot com>]

This is a re-posting of a patch.

To recap: store_expr does a little optimization which removes a copy
insn.  This is usually OK, but if the right hand side of a store expr
can trap this is wrong.

This is the case I'm interested in:

   a = ({ tmp = foo->bar; tmp; });

We must not convert this into

   a = foo->bar;

I'm not sure why we're doing this optimization during store_expr,
because the temporary will be deleted by a later compiler pass.  I
suppose doing the job early reduces the amount of RTL that we process
and thus speeds up the compiler.

In any case, disabling this optimization when using non-call
exceptions should only make a difference to optimized code when it
would otherwise be incorrect.

If anyone can think of a better way to fix this I'm all ears.

Andrew.

	* expr.c (store_expr): Always return target if we're using
	non-call exceptions and optimizing.

Index: expr.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/expr.c,v
retrieving revision 1.494
diff -c -2 -p -r1.494 expr.c
*** expr.c	27 Nov 2002 02:29:11 -0000	1.494
--- expr.c	10 Dec 2002 17:14:29 -0000
*************** store_expr (exp, target, want_value)
*** 4454,4458 ****
  	  && !(GET_CODE (target) == MEM && MEM_VOLATILE_P (target))
  	  && ! rtx_equal_p (temp, target)
! 	  && (CONSTANT_P (temp) || want_value))
  	dont_return_target = 1;
      }
--- 4454,4459 ----
  	  && !(GET_CODE (target) == MEM && MEM_VOLATILE_P (target))
  	  && ! rtx_equal_p (temp, target)
! 	  && (CONSTANT_P (temp) || want_value)
! 	  && !(flag_non_call_exceptions && optimize))
  	dont_return_target = 1;
      }

Andrew Haley writes:
 > I have found another case when using non-call exceptions where a
 > temporary variable is incorrectly elided.  In this case, expanding a
 > store expr turns this:
 > 
 >    a = ({ tmp = foo->bar; tmp; });
 > 
 > into this:
 > 
 >    a = foo->bar;
 > 
 > Andrew.
 > 
 > 	* expr.c (store_expr): Always return target if we're using
 > 	non-call exceptions.
 > 
 > Index: expr.c
 > ===================================================================
 > RCS file: /cvs/gcc/gcc/gcc/expr.c,v
 > retrieving revision 1.488
 > diff -u -r1.488 expr.c
 > --- expr.c	15 Oct 2002 20:09:32 -0000	1.488
 > +++ expr.c	19 Nov 2002 17:31:41 -0000
 > @@ -4398,7 +4398,8 @@
 >  	    && REGNO (target) < FIRST_PSEUDO_REGISTER)
 >  	  && !(GET_CODE (target) == MEM && MEM_VOLATILE_P (target))
 >  	  && ! rtx_equal_p (temp, target)
 > -	  && (CONSTANT_P (temp) || want_value))
 > +	  && (CONSTANT_P (temp) || want_value)
 > +	  && !flag_non_call_exceptions)
 >  	dont_return_target = 1;
 >      }
 >