Re: reload: fix for gcc.c-torture/execute/990127-1.c -O0 on SH

[Richard Henderson <rth at cygnus dot com>]

On Mon, Aug 21, 2000 at 06:12:24PM -0600, Jeffrey A Law wrote:
> How can we compute accurate death notes for the frame pointer before reload
> has completed?  Any reference to a pseudo would need to make the FP live
> since the pseudo might not get a hard reg and would be loaded from the
> current frame.

Fixed thus.  I checked that it fixes Alex's original problem as well.


r~


        * flow.c (calculate_global_regs_live): Mark frame pointer live
        everywhere before reload.

Index: flow.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/flow.c,v
retrieving revision 1.327
diff -c -p -d -r1.327 flow.c
*** flow.c	2000/08/21 09:18:05	1.327
--- flow.c	2000/08/22 00:42:42
*************** calculate_global_regs_live (blocks_in, b
*** 3191,3196 ****
--- 3191,3202 ----
  	 the case for blocks within infinite loops.  */
        SET_REGNO_REG_SET (new_live_at_end, STACK_POINTER_REGNUM);
  
+       /* Similarly for the frame pointer before reload.  Any reference
+ 	 to any pseudo before reload is a potential reference of the
+ 	 frame pointer.  */
+       if (! reload_completed)
+ 	SET_REGNO_REG_SET (new_live_at_end, FRAME_POINTER_REGNUM);
+ 
        /* Regs used in phi nodes are not included in
  	 global_live_at_start, since they are live only along a
  	 particular edge.  Set those regs that are live because of a