[llvm-commits] [llvm-gcc-4.2] r117934 - /llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp

[Jim Grosbach]

On Nov 1, 2010, at 8:52 PM, Eric Christopher wrote:

> 
> On Nov 1, 2010, at 10:37 AM, Bob Wilson wrote:
> 
>> I don't know so much about synchronization on other targets besides ARM, and
>> since there was already a separate version of the barriers for ARM/Darwin,
>> I've only changed that version.  It seems like this change would apply to
>> other targets as well, but I don't know for sure.
> 
> x86 largely uses the lock prefix to handle this sort of thing as opposed to emitting
> a memory barrier instruction (mfence, lfence, sfence).
> 

Yeah, the X86 atomic stuff explicitly removes the membarrier intrinsics when they surround one of the atomic operation intrinsics.

> This change in particular isn't real interesting for x86 since we'll still just use an
> mfence there I think. :)
> 
> -eric
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits