cvs commit: src/sys/conf kern.mk

[Marius Strobl]

On Mon, Dec 04, 2006 at 02:33:24PM -0800, Kip Macy wrote:
> I haven't talked to Sam, I had forgotten about that - this will only break
> ABI compat on -HEAD.  "-mcmodel=medlow" was almost certainly never needed,
> its an optimization that reduces then number of instructions needed to load
> a constant by assuming that the constant is <=32 bits. Loading a 32 bit
> constant is 3 instructions,  loading a 64-bit constant is 7 :-(. Considering
> the other performance issues that sparc64 has, its safe to say that the
> extra overhead is in the noise.
> 
>                                                          -Kip

I just remember commits like sys/conf/Makefile.sparc64 rev.
1.14 and some others which claim that model medlow is/was
essential, unfortunately neither did bother to explain as to
why that's supposed to be so...
I suppose that using medany also fixes executing modules
on sparc64/sun4u machines with >4GB RAM, correct?

Marius

> 
> 
> On 12/4/06, Marius Strobl <marius at alchemy.franken.de> wrote:
> >
> >On Sun, Nov 26, 2006 at 11:16:46PM +0000, Kip Macy wrote:
> >> kmacy       2006-11-26 23:16:46 UTC
> >>
> >>   FreeBSD src repository
> >>
> >>   Modified files:
> >>     sys/conf             kern.mk
> >>   Log:
> >>   Fix module loading for sparc on machines with greater than 1GB of RAM
> >>
> >
> >Hi Kip,
> >
> >could you please update the comment above the line you changed
> >as appropriate?
> >Do you have an idea why medlow no longer is crucial in order
> >to get a working kernel?
> >Did you talk to sam@ regarding providing a new sparc64 HAL BLOB
> >(according to sys/contrib/dev/ath/public/sparc64-be-elf.inc it
> >was built with -mcmodel=medlow, too)?
> >
> >Marius
> >
> >