[dataflow]: PATCH: ia64 bootstrap and regression fixes

Seongbae Park seongbae.park@gmail.com
Tue Nov 28 17:09:00 GMT 2006


On 11/28/06, Kenneth Zadeck <zadeck@naturalbridge.com> wrote:
> For now this is ok.  I want you to think about not adding those extra
> bit vectors to the lr problem.  These are only used for register
> allocation and will add a lot of space and time to the rest of the

I agree. I tried to avoid using more space in a couple of different ways
but in the end, I've concluded that adding TOP is the only way forward,
as long as there are forward scanning passes (local/global/reload)
that rely on the exact liveset.

> compilation.  The urec problem is only used for register allocation and
> it seems like if you can put the vectors and the computation there, you
> can save a lot of time and space.

It might be possible to compute IN set on the fly based on TOP.
ADEF and AUSE are allocated on-demand, exactly because I wanted to
avoid wasting space since adef and ause are empty most of the time.
-- 
#pragma ident "Seongbae Park, compiler, http://seongbae.blogspot.com"



More information about the Gcc-patches mailing list