This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: flow bug


On Mon, Mar 08, 1999 at 11:04:02PM -0700, Jeffrey A Law wrote:
> debug_rtx_list (NEXT_INSN (b->end), 100) shows:
> 
> (code_label 10 9 20 3 ("101"))
> 
> (note 20 10 15 [bb 1] NOTE_INSN_BASIC_BLOCK)
> 
> (note 15 20 0 0 NOTE_INSN_BLOCK_END)
> 
> So next_nonnote_insn (q) returns null.  Then we try to examine the code
> of Q, which causes the segfault.

I think as well we need the same cc0 check that you put in for
merge_blocks_nomove.  

I'm gonna bootstrap several machines before I put this in.


r~



	* flow.c (tidy_fallthru_edge): Be more careful finding the last
	BARRIER of a list.  Delete the cc0 setter as well as a cond jump.

Index: flow.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/flow.c,v
retrieving revision 1.104
diff -c -p -d -r1.104 flow.c
*** flow.c	1999/03/08 14:40:47	1.104
--- flow.c	1999/03/09 08:08:07
*************** tidy_fallthru_edge (e, b, c)
*** 1933,1939 ****
       edge e;
       basic_block b, c;
  {
!   rtx p, q, h;
  
    /* ??? In a late-running flow pass, other folks may have deleted basic
       blocks by nopping out blocks, leaving multiple BARRIERs between here
--- 1933,1939 ----
       edge e;
       basic_block b, c;
  {
!   rtx q, h;
  
    /* ??? In a late-running flow pass, other folks may have deleted basic
       blocks by nopping out blocks, leaving multiple BARRIERs between here
*************** tidy_fallthru_edge (e, b, c)
*** 1943,1954 ****
       barriers and notes.  */
  
    q = NEXT_INSN (b->end);
!   do
!     {
!       p = q;
!       q = next_nonnote_insn (q);
!     }
!   while (GET_CODE (q) == BARRIER);
  
    /* Assert that we now actually do fall through.  */
    h = c->head;
--- 1943,1952 ----
       barriers and notes.  */
  
    q = NEXT_INSN (b->end);
!   if (q && GET_CODE (q) == NOTE)
!     q = next_nonnote_insn (q);
!   while (q && GET_CODE (q) == BARRIER)
!     q = next_nonnote_insn (q);
  
    /* Assert that we now actually do fall through.  */
    h = c->head;
*************** tidy_fallthru_edge (e, b, c)
*** 1963,1968 ****
--- 1961,1973 ----
    q = b->end;
    if (GET_CODE (q) == JUMP_INSN)
      {
+ #ifdef HAVE_cc0
+       /* If this was a conditional jump, we need to also delete
+ 	 the insn that set cc0.  */
+       if (! simplejump_p (q) && condjump_p (q))
+ 	q = PREV_INSN (q);
+ #endif
+ 
        if (b->head == q)
  	{
  	  PUT_CODE (q, NOTE);
*************** tidy_fallthru_edge (e, b, c)
*** 1974,1980 ****
      }
  
    /* Selectively unlink the sequence.  */
!   delete_insn_chain (NEXT_INSN (q), p);
  
    e->flags |= EDGE_FALLTHRU;
  }
--- 1979,1986 ----
      }
  
    /* Selectively unlink the sequence.  */
!   if (q != PREV_INSN (c->head))
!     delete_insn_chain (NEXT_INSN (q), PREV_INSN (c->head));
  
    e->flags |= EDGE_FALLTHRU;
  }


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]