romcc-0.36 with -O2 8 registers are sufficient

I just got to looking at my test cases and I have successfully
compiled all of my test cases that are derivatives of actual
memory initialization code with -O2 and no extra registers.

Which is a very good sign for the case of using extra registers
to avoid inlining subroutines.

Eric

[Stefan Reinauer]

* Eric W. Biederman  [031011 20:38]:
> I just got to looking at my test cases and I have successfully
> compiled all of my test cases that are derivatives of actual
> memory initialization code with -O2 and no extra registers.

It also translates the K8 smbus_write_byte() function now.. great!
> Which is a very good sign for the case of using extra registers
> to avoid inlining subroutines.

Does this mean we can push the size of the linuxbios image down 
noticably now?

Stefan

Stefan Reinauer  writes:
> Eric W. Biederman  [031011 20:38]:
> > I just got to looking at my test cases and I have successfully
> > compiled all of my test cases that are derivatives of actual
> > memory initialization code with -O2 and no extra registers.
>
> It also translates the K8 smbus_write_byte() function now.. great!
> > Which is a very good sign for the case of using extra registers
> > to avoid inlining subroutines.
>
> Does this mean we can push the size of the linuxbios image down 
> noticably now?

That is the goal.  .36 allows static const variables outside
of functions, and does not duplicate them when they are inside
of functions.

-O2 helps a little bit.

I am just starting on not inlining function calls.  Once that
is working we can see if there is noticeable size decrease.  I expect
that to be the case, but you never know until you try.

Eric