Re: Altivec patches for glibc

[Konstantinos Margaritis <markos at debian dot gr>]

On ÎÏÎ 16 ÎÎÎ 2004 23:56, Steve Munroe wrote:
> You also have to address a world where GLIBC is shared with many
> PowerPC implementations, most of which don't have Altivec (VMX)
> extensions. So what ever changes are made glibc has to continue to
> run correctly on POWER3/POWER4/POWER5 and many imbedded systems.
> This applies to both 32- and 64-bit ABIs.

Of course, there will be (or there already available in glibc 
actually) both compile time and runtime checks for Altivec. We 
understand that it is equally important to keep compatibility. The 
check for Altivec can be done once and glibc will offer the correct 
routines as appropriate. It's definitely not a question of removing 
existing scalar code, just allowing for optimized routines to be used 
if the capability is there.

> There is also the complex issue of performance comparison which
> gets tricky when a single code base is shared across processors
> from 4XX imbedded chips to 64-Way POWER5 systems. Comparing hand
> coded VMX to the code generated by current (GCC-3.3) GCC can lead
> to false conclusions. The code generation issue is also changing
> with GCC-4.x which will support more aggressive optimization in
> addition to auto vectorization. And finally the relative advantage
> of VMX in reduced in 64-bit mode by comparison to 32-bit.

AFAIK, there has never been made a measurement of such a widespread 
use of Altivec in such a vital component of the OS. This was the 
incentive to do it, perhaps the performance it total will not be 20x 
(of course I don't expect it to be that much) but one cannot really 
say for sure, just how much better will eg. kernel compile times will 
be. Until we do such a benchmark, I don't think we can really say for 
sure what the benefits will be. Still, I at least expect a 
substantial gain from it.
Of course, before/if the patches will make it to glibc itself, I 
intend to do some heavy testing on this.

Konstantinos