Re: Loop clean up

[Michael Hayes <m dot hayes at elec dot canterbury dot ac dot nz>]

Clinton Popetz writes:

 > This is a problem with Michael's original loop patch.  See my recent
 > mesg under "[Re: [EGCS 2.96, 2000/01/17 snapshot] Internal compiler
 > error in loop_iterations" or at:
 > 
 > 	http://gcc.gnu.org/ml/gcc-bugs/2000-01/msg00475.html


Yes, I somehow didn't slurp up Bernd's patch and never noticed the
regression since my vanilla source tree had the same bug.  My
apologies.

After working through the problem, I think the following should be a
better solution to what was proposed by Bernd.  I do not notice any
new regressions with this.  Shall I commit it to right the ship?

Michael.


Index: unroll.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/unroll.c,v
retrieving revision 1.87
diff -c -3 -p -r1.87 unroll.c
*** unroll.c	2000/01/17 15:55:18	1.87
--- unroll.c	2000/01/19 05:21:56
*************** loop_iterations (loop)
*** 3702,3711 ****
        return 0;
      }
  
!   /* The only new registers that care created before loop iterations are
!      givs made from biv increments, so this should never occur.  */
  
!   if ((unsigned) REGNO (iteration_var) >= reg_iv_type->num_elements)
      abort ();
  
    iteration_info (iteration_var, &initial_value, &increment,
--- 3702,3715 ----
        return 0;
      }
  
!   /* The only new registers that are created before loop iterations
!      are givs made from biv increments or registers created by
!      load_mems.  In the latter case, it is possible that try_copy_prop
!      will propagate a new pseudo into the old iteration register but
!      this will be marked by having the REG_USERVAR_P bit set.  */
  
!   if ((unsigned) REGNO (iteration_var) >= reg_iv_type->num_elements
!       && ! REG_USERVAR_P (iteration_var))
      abort ();
  
    iteration_info (iteration_var, &initial_value, &increment,