www/firefox on NetBSD/m68k ?

Hi,

> Never did hear if you got FF to build on your Amiga. 
How'd that go?

There were many issues which kept me from running NetBSD 3.
I think 
they're fixed now, and I'm rebuilding NetBSD 3 from
scratch (there are a 
LOT of emulated instructions being processed on the m68060s,
which slows 
things down dramatically), after which I'll restart the
bulk building. 
I'll post status when that's running.

John Klos
-- 
Today is the day that my destiny calls me!
   Meowth

John Klos writes: 

>  I think 
> they're fixed now, and I'm rebuilding NetBSD 3 from
scratch (there are a 
> LOT of emulated instructions being processed on the
m68060s, which slows 
> things down dramatically), 

How big a penalty are the emulated instructions on the 68060
in practice?  
I've been thinking of digging out my mvme177s. 

>>  I think they're fixed now, and I'm rebuilding
NetBSD 3 from scratch (there 
>> are a LOT of emulated instructions being processed
on the m68060s, which 
>> slows things down dramatically), 
>
> How big a penalty are the emulated instructions on the
68060 in practice? 
> I've been thinking of digging out my mvme177s.

Pretty darn big. It causes an exception and all of the
overhead associated 
with exception handling PER INSTRUCTION. I think that's why
-current is 
painfully slow on the m68060s, although I didn't have
enough patience to 
wait for a full boot to see if that was really it.

On the other hand, after recompiling the whole OS with
-m68060, the 
performance should be as good as, or better, than with older
NetBSD, and 
MUCH better than the daily NetBSD 3 binaries.

An mvme177 should be pretty quick with NetBSD...

John Klos
-- 
Today is the day that my destiny calls me!
   Meowth

John Klos wrote:
> On the other hand, after recompiling the whole OS with
-m68060, the 
> performance should be as good as, or better, than with
older NetBSD, and 
> MUCH better than the daily NetBSD 3 binaries.

What kind of gains does one see optimizing for a specific
processor as 
opposed to a processor family?  Just curious.


Tim

-- 
Tim & Alethea
christtrek.org

On Wed, Aug 09, 2006 at 07:32:48AM -0500, Tim & Alethea
Larson wrote:
> What kind of gains does one see optimizing for a
specific processor as 
> opposed to a processor family?  Just curious.

That depends a lot on the processors & families
involved.  In a case
such as this where the 68060 has to handle some 68k
instructions by
emulation, optimizing for the 68060 (and avoiding those
instructions)
can make a huge difference.  In some other cases where some
instructions
or instruction sequences are more efficient on a given
processor and gcc
is aware of that, you'll get some gain--but the actual gain
depends on
how much the code can be affected by those differences.  In
long-lived
processor families, there are some pretty wide differences
in what each
processor does well.

I expect that there's some literature out there on it, but
it's not
an area I've studied.

-allen

-- 
Allen Briggs  |  http://www.ninthw
onder.com/~briggs/  |  briggsninthwonder.com

On Wed, Aug 09, 2006 at 07:32:48AM -0500, Tim & Alethea
Larson wrote:
> John Klos wrote:
> >On the other hand, after recompiling the whole OS
with -m68060, the 
> >performance should be as good as, or better, than
with older NetBSD, and 
> >MUCH better than the daily NetBSD 3 binaries.
> 
> What kind of gains does one see optimizing for a
specific processor as 
> opposed to a processor family?  Just curious.

-m68060 does avoid the 32bit time 32 bit with 64 bit result
multiplication
and the 64 bit by 32 bit division with 32 bit result.

Does gcc create them nowadays more often than it used to?

Btw: the NetBSD/amiga kernel is always compiled with
-m68060, maybe
unless 68060 support is not compiled into the kernel at
all...

	-is
-- 
seal your e-mail: http://www.gnupg.org/