On Monday, August 4, 2003, at 07:51 AM, Jason Thorpe wrote:

>
> On Sunday, August 3, 2003, at 06:31  PM, Simon Burge wrote:
>
>> Does this have any measurable impact on trap response and context 
>> switch
>> times (simply because of more instruction fetches)?  It's also larger
>> code, so we have greater I$ pollution...
>
> It might be worth #ifdef'ing the use of those insns, using 
> __OPTIMIZE_SIZE__ to use the load/store-multiple insns on non-PPC64.

I thought of that.  But use of stmw/lmw are usually buried inside other 
macros and makes even more messy.  The amount of instructions is 
relatively minor
and is (IMO) not worth the effort.

> It's worth noting, however, that individual load/store insns are 
> probably going to be faster than the load/store-multiple on most 
> modern PPCs (like, probably everything since the 601).

That has been my experience.
-- 
Matt Thomas                     email: matt@3am-software.com
3am Software Foundry              www: 
http://3am-software.com/bio/matt/ Cupertino, CA              
disclaimer: I avow all knowledge of this message.