Fix buglet in previous mechanical loop.c change

Fix buglet in previous mechanical loop.c change

[Jeffrey A Law]

[-- Attachment #1: Type: text/plain, Size: 1600 bytes --]

The recent change to use various predicates instead of explicitly
testing for suitable rtx codes had a minor flaw.  Consider this
change fragment from loop.c:

***************
*** 5482,5495 ****
      }
   
    /* Look for givs which are memory addresses.  */
!   if (GET_CODE (p) == INSN)
      find_mem_givs (loop, PATTERN (p), p, not_every_iteration,
                   maybe_multiple);
   
    /* Update the status of whether giv can derive other givs.  This can
       change when we pass a label or an insn that updates a biv.  */
!   if (GET_CODE (p) == INSN || GET_CODE (p) == JUMP_INSN
!       || GET_CODE (p) == CODE_LABEL)
      update_giv_derive (loop, p);
    return p;
  }
--- 5479,5491 ----
      }
   
    /* Look for givs which are memory addresses.  */
!   if (NONJUMP_INSN_P (p))
      find_mem_givs (loop, PATTERN (p), p, not_every_iteration,
                   maybe_multiple);
   
    /* Update the status of whether giv can derive other givs.  This can
       change when we pass a label or an insn that updates a biv.  */
!   if (INSN_P (p))
      update_giv_derive (loop, p);
    return p;
  }


In particular note that we no longer call update_giv_derive if P is a
CODE_LABEL.

This can cause the loop optimizer to incorrectly reduce a GIV when it's
BIV is conditionally incremented.

I've been unable to trip this with unmodified sources, but it caused a
bootstrap failure (hang building libgcc with the stage2 compiler) with
a hacked compiler I'm using to evaluate the jump threading code.

This patch restores the previous behavior.

Bootstrapped on i686-pc-linux-gnu.



[-- Attachment #2: ZZZ --]
[-- Type: text/x-patch, Size: 1402 bytes --]

Index: ChangeLog
===================================================================
RCS file: /cvs/gcc/gcc/gcc/ChangeLog,v
retrieving revision 2.4542
diff -c -p -r2.4542 ChangeLog
*** ChangeLog	15 Jul 2004 13:46:05 -0000	2.4542
--- ChangeLog	15 Jul 2004 14:52:46 -0000
***************
*** 1,3 ****
--- 1,8 ----
+ 2004-07-15  Jeff Law  <law@redhat.com>
+ 
+ 	* loop.c (check_insn_for_givs): Restore check for code labels that was
+ 	accidentally deleted by a recent checkin.
+ 
  2004-07-15  Nathan Sidwell  <nathan@codesourcery.com>
  
  	* vec.h (VEC_T_truncate): Allow truncation of an empty vector.
Index: loop.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/loop.c,v
retrieving revision 1.503
diff -c -p -r1.503 loop.c
*** loop.c	9 Jul 2004 03:29:33 -0000	1.503
--- loop.c	15 Jul 2004 14:53:01 -0000
*************** check_insn_for_givs (struct loop *loop, 
*** 5485,5491 ****
  
    /* Update the status of whether giv can derive other givs.  This can
       change when we pass a label or an insn that updates a biv.  */
!   if (INSN_P (p))
      update_giv_derive (loop, p);
    return p;
  }
--- 5485,5491 ----
  
    /* Update the status of whether giv can derive other givs.  This can
       change when we pass a label or an insn that updates a biv.  */
!   if (INSN_P (p) || LABEL_P (p))
      update_giv_derive (loop, p);
    return p;
  }

Re: Fix buglet in previous mechanical loop.c change

[Daniel Jacobowitz]

On Thu, Jul 15, 2004 at 08:55:01AM -0600, Jeffrey A Law wrote:
> The recent change to use various predicates instead of explicitly
> testing for suitable rtx codes had a minor flaw.  Consider this
> change fragment from loop.c:
> 
> !   if (GET_CODE (p) == INSN || GET_CODE (p) == JUMP_INSN
> !       || GET_CODE (p) == CODE_LABEL)

> !   if (INSN_P (p))

> In particular note that we no longer call update_giv_derive if P is a
> CODE_LABEL.
> 
> This can cause the loop optimizer to incorrectly reduce a GIV when it's
> BIV is conditionally incremented.

And you changed this to:

> !   if (INSN_P (p) || LABEL_P (p))

This now accepts CALL_INSN, which it didn't used to.  Was that a bug in
the old code?

-- 
Daniel Jacobowitz

Re: Fix buglet in previous mechanical loop.c change

[Jeffrey A Law]

On Thu, 2004-07-15 at 09:11, Daniel Jacobowitz wrote:
> On Thu, Jul 15, 2004 at 08:55:01AM -0600, Jeffrey A Law wrote:
> > The recent change to use various predicates instead of explicitly
> > testing for suitable rtx codes had a minor flaw.  Consider this
[ ... ]

> 
> This now accepts CALL_INSN, which it didn't used to.  Was that a bug in
> the old code?
Acceping a CALL_INSN is conservatively correct.  It won't have any
effect on the behavior of update_giv_derive except that unnecessary
work may be done.

Jeff