[PATCH] Disable tree-stdarg on a per-target basis

Giovanni Bajo rasky@develer.com
Thu May 26 18:32:00 GMT 2005


Eric Botcazou <ebotcazou@libertysurf.fr> wrote:

>>> true on SPARC (I think the RTL CSE + DCE passes already eliminate as
much
>>> stdarg overhead as tree-stdarg would there).
>>
>> Do we want to keep on relying on the RTL optimizers for this?  Moving
this
>> optimization up to the tree level is on par with the current trend, even
if
>> there is no immediate benefit in code generation.
>
> Not only there is no immediate benefit in code generation (unless I'm
> mistaken), but there is immediate slowdown because of an additional pass.
> Given that GCC is already not the fastest compiler on Earth (on SPARC too,
the
> Sun Studio compiler outperforms it),

We added more than 70 passes between 3.4 and 4.0, and we are not performing
that bad. Are you saying that the tree stdarg pass causes a slowdown which
is really measurable by itself?

> I think my position is not unreasonable.

Sure, I was just trying to understand why.
-- 
Giovanni Bajo



More information about the Gcc-patches mailing list