Re: REGNO_DECL cleanup

[Jan Hubicka <jh at suse dot cz>]

> On Mon, Dec 23, 2002 at 01:34:38AM +0100, Jan Hubicka wrote:
> > 	* emit-rtl.c (reg_attrs_htab): New static variable.
> > 	(reg_attrs_htab_hash, reg_attrs_htab_eq, get_reg_attrs): New static
> > 	functions.
> > 	(reg_rtx): Do not maintain regno_decl.
> > 	(gen_rtx_REG_offset, set_reg_attrs_from_mem, set_delc_rtx,
> > 	set_mem_attrs_from_reg): New global function.
> > 	(init_emit): Do not initialize regno_decl.
> > 	(init_emit_once): initialize reg_attrs_htab.
> > 	* final.c (alter_subreg): Do not replace REG by SUBREG.
> > 	(gen_mem_expr_from_op): Improve output.
> > 	(output_asm_operands): Likewise.
> > 	* function.c (assign_params): Do not set REGNO_DECL.
> > 	* function.h (struct function): Kill regno_decl.
> > 	(REGNO_DECL): Kill.
> > 	* gengtype.c (adjust_field_rtx_def): Handle new field of reg.
> > 	* print_rtl.c (print_rtx): Output REG information.
> > 	* regclass.c (reg_scan_mark_refs): Update attrs.
> > 	* reload1.c (alter_reg): Likewise.
> > 	* simplify_rtx.c (simplify_subreg): Likewise.
> > 	* stmt.c (expand_decl): Likewise.
> > 	* rtl.def (REG): Add new field.
> > 	* rtl.h (struct reg_attrs): New.
> > 	(rtunion_def): At rtreg.
> > 	(X0MEMATTR): Add checking.
> > 	(X0REGATTR, REG_ATTRS, REG_EXPR, REG_OFFSET): New macro.
> > 	(set_reg_attrs_from_mem, set_mem_attrs_from_reg, gen_rtx_REG_offset):
> > 	Declare.
> > 	* tree.h (SET_DECL_RTL): Call set_decl_rtl.
> 
> Ok.

Thanks a lot!  Thius would allow to continue working on location lists
support for DWARF2.

Josef has fixed compile time problems with variable tracking and by
looking briefly at the output it seems to work quite well now.
I will try to update the dwarf2 bits this weekend.  Problem is that
still GDB part is missing as we are in the kind of deadlock right now.
Do you think it is good idea to merge the bits into mainline GCC or
should we wait for GDB support to become available?
(in the second I would probably concentrate on merging some other bits
from the branch first)

Honza
> 
> 
> r~