[Debian-ppc64-devel] I am planning to restart ppc64 ports project.

[Thibaut VARÈNE]

Le 4 oct. 10 à 14:43, Hiroyuki Yamamoto a écrit :

>> I'm sitting on one of those ;P). Again, 64-bit code performs 15 to  
>> 25%
>> slower on ppc processors than the equivalent 32-bit code. 64-bit code
>> on power processors has very limited usefulness, only for some very
>> specific workloads. At the very least, a complete 64-bit distribution
>> would only provide a huge penalty loss.
>
> I know it.
> I understand that the former ppc64 porting work was canceled because  
> of it.
> Therefore, I am hesitating in the ppc64 porting direction for VMX-128,
> though this option is not possible to specify now.

I'm not sure what you want to do. VMX128 and PPC64 are two different  
things. VMX128 is IBM efforts to enhance altivec for the Xenon. I'm  
not aware of any other use case. Therefore your potential target is  
extremely reduced (who is running Linux on their Xbox360?). The  
probability of it gaining interest from the community is therefore  
quite low...

> (The ppc64 port should be led for high-end servers?)

High-end servers users will - again - certainly not want to shave 20%  
average performance off their supercomputers. There's a reason why  
*nobody* (and that includes high-end servers users) ever cared about  
Debian PPC64...

> I think that such experiments should be continued for the future.

Aside the academic purpose and personal curiosity satisfaction, it's  
quite pointless, TBH.

HTH

-- 
Thibaut VARÈNE
http://www.parisc-linux.org/~varenet/