[patch] Shark reroute LLVM atomic intrinsics to Zero

[Xerxes Rånby]

Robert Schuster skrev:
> Hi Andrew,
>
> Andrew Haley schrieb:
>   
>> Definitely; we really don't want a function call just do do an atomic
>> cmpxchg.
>>     
> That is why I opted for enabling it only for the arches that really need
> it by using #ifdef.
>
>   
Before I start prepare a patch adding lots of #ifdefs all over 
sharkBuilder for all kinds of architectures, i would like to ask if it 
would be better to do this instead:
I consider implementing a new configure option say 
--enable-shark-atomic-reroute that can enable or disable this rerouting 
thus then we only needs to add one #ifdef SHARK_LLVM_ATOMIC_REROUTE  
around the parts that i patch in.
My thinking is that it should be easy to add new zero/shark 
architectures without having to update sharkBuilder at all.
configure should be made to set this to be enabled for all zero 
architectures that we know needing it.

>>  This is really just a workaround for an llvm bug, and hopefully
>> it'll soon go away.
>>     
> Do you know more than me?
>
> The 'bug' is actually a missing feature and I don't see someone working
> on this. I tried fixing it myself but it will take me a while until
> understand how to use llvm's tablegen syntax to describe the generation
> of a series of instructions for the compiler
Found this great google-find: http://markmail.org/message/73owc5nrvsbmrhes
It is the mail correspondence when Gary implemented the atomic 
intrinsics in llvm for PPC.
Great work Gary!

Cheers, and have a great day.
Xerxes