[Clfs-dev] Optional packages for graphite

[Martin Ward]

On 26/07/13 20:56, William Harrington wrote:
>
> On Jul 24, 2013, at 1:36 AM, William Harrington wrote:
>
>> I"m also curious, do we need to build ISL and CLooG in cross-tools? 
>> It seems best that optimized binaries would be using the graphite 
>> toolchain from tools and then final-system. I see no need to build 
>> graphite in the cross-tools. It would make sense to enable graphite 
>> in the tools, and definitely in final-system.
>
> Okay, I've verified, using x86 for a powerpc target that it isn't 
> possible to build a temp system with graphite options.
>
> I do not know if graphite options can be used when cross compiling, 
> but it seems it can't be. Now, it may work fine when the host system 
> is the same arch as the target system, such as ppc to ppc or x86 to 
> x86, however, from x86 to ppc it seems graphite options can't be enabled.
>
> A message during configure after an attempted compiler check:
>
> sorry, unimplemented: Graphite loop optimizations cannot be used 
> (-fgraphite, -fgraphite-identity, -floop-block, -floop-interchange, 
> -floop-strip-mine, -floop-parallelize-all, and -ftree-loop-linear
>
> I have more testing to do. But if this is for sure exact, then we do 
> not need cloog and isl in cross-tools or tools.
>
> Now, the only reason to use it in tools, is when starting the final 
> system and someone wants to use graphite options before rebuliding gcc.
>
> Can anyone else confirm this?
>
> Sincerely,
>
> WIlliam Harrington
>
>
> _______________________________________________
> Clfs-dev mailing list
> Clfs-dev at lists.cross-lfs.org
> http://lists.cross-lfs.org/listinfo.cgi/clfs-dev-cross-lfs.org
I suspect that graphite optimisation is in it's early stages with lots 
of bugs and oddities as shown here,
http://forums.gentoo.org/viewtopic-t-964490.html?sid=102e752fb6f57b6c9dbf095d05548644

but i suspect it will become a necessary dependency rather than 
optional  at some stage

it might be prudent to keep it in the build just to be up to speed if 
there is a change but no hard feeling whether is should be in or out

Martin