Re: New Regalloc Branch

[Denis Chertykov <denisc at overta dot ru>]

Erwin Ingris <erwiningris@yahoo.com.au> writes:

> Hi,
>  
> > The current allocator (which really works in three
> > passes) doesn't work
> > really globally, and the algorithms it uses are
> > failry adhoc.  It can't
> > split pseudo registers, or coalesce them (although
> > it has something like
> > preferences to get rid of copies, which is something
> > like coalescing).
> > It doesn't use exact live-ness information to the
> > full extent. And
> > spill-code is inserted by a different pass (reload),
> 
> In reg-alloc branch code I am seeing a call to
> reload after reg_alloc.
> 
>  3090   if (flag_new_regalloc)
>    3091     {
>    3092       delete_trivially_dead_insns (insns,
> max_reg_num ());
>    3093       reg_alloc ();
>    3094
>    3095       timevar_pop (TV_LOCAL_ALLOC);
>    3096       if (dump_file[DFI_lreg].enabled)
>    3097         {
>    3098           timevar_push (TV_DUMP);
>    3099
>    3100           close_dump_file (DFI_lreg, NULL,
> NULL);
>    3101           timevar_pop (TV_DUMP);
>    3102         }
>    3103
>    3104       /* XXX clean up the whole mess to bring
> live info in shape
>          again.  */
>    3105       timevar_push (TV_GLOBAL_ALLOC);
>    3106       open_dump_file (DFI_greg, decl);
>    3107
>    3108       build_insn_chain (insns);
>    3109       failure = reload (insns, 0);
> 
> would not that imply spill code in a seperate pass
> after reg_alloc or I am missing something.

Reload used for following things:
1. Eliminate registers (fp elimination, argp elimination, ...);
2. Add registers for (match_scratch ...);
3. Fixing wrong insns, Insns with constrains '%0...9';
4. Fixing wrong addresses.
5. Old (original) regclass can choose wrong class - class which will not
   satisfy constraints from all insns in which register used;
6. May be I forgot something.

> As I understand handling spill should be part
> of reg_alloc from ur previous reply.

Handling spill is a part of reg_alloc.


Denis.