This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch to use push instead of sub
> On Sat, Apr 17, 1999 at 04:04:28PM +0200, Jan Hubicka wrote:
> > This patch implements this. It use split to translate addsi3 pattern to
> > push, so scheduler is aware of this change.
>
> One problem with this split is that it introduces an artificial
> scheduling dependancy on eax. Better would be to use a pattern like
>
> (set (mem:SI (pre_dec:SI (reg:SI 7)))
> (unspec:SI [(const_int 0)] 9))
Huh. I am not sure why you think this is better. outputed instruction "push eax"
is better then "push something else" so i think we will want to keep it.
This instruction have real dependency on eax, so I believe it is good to tell about this to s heduler.
>
> Related to this and your other pending AGI patch, have_esp_bypass is
> getting more complicated than it needs to be. I would recommend adding
> "push" and "pop" to attribute "type" and using them appropriately.
Yes,.. I was thinking about new have_esp_bypass attribute and handle the
ret/call insns using this as well. I believe this is a better choice,
so I will come with united patch for this two changes with have-esp_bypass attribute.
Honza