Re: mips64 on port-cobalt

["Alex Pelts" <alexp%broadcom.com@localhost>]

On 12/18/2009 1:58 PM, Andy Ruhl wrote:
> On Fri, Dec 18, 2009 at 9:54 AM, Alex Pelts <alexp%broadcom.com@localhost> 
> wrote:
>> Not that this should stop anyone from trying, but I am not sure Qube
>> will gain anything by going 64 bit. While RM5231 processor is certainly
>> 64bit, the system interface is 32bit so for every data memory read and
>> write it will require 2 accesses in 64 bit mode. Also data cache will be
>> effectively cut in half if most of the structure size would double due
>> to going to 64 bit.
>> Even all of this is mediated somehow, context switch will become 2x
>> slower due to doubling amount of data that needs to be saved and restored.
>> So aside from satisfying curiosity the 64bit mode will be detrimental to
>> Qube performance. And this thing already is not a speed daemon. Plus
>> this system can't physically accommodate enough memory to need 64 bits.
> 
> Sometimes when you're doing research and trying new stuff you learn
> something you didn't expect.
> 
> I'm not sure what the chances of that are here, probably not super duper good.
> 
> I always thought the whole point of working with this wacky stuff was
> to make the MI code stronger.
> 
> Andy
> 

I do not disagree with anything that you said. This is just a note for
people who might expect a performance boost from 64bit implementation.
Unless you are doing some math that involves 64bit numbers somehow, I
would expect general slowdown of things.

I was curious how people who owns Qubes rebuild their packages. I kind
of gave up on my qube after it started to take me few days to re-build
handful of packages that I use.

Regards,
Alex