[llvm-commits] Patch: address space support for memcpy/memmove/memset

[Chris Lattner]

On Mar 29, 2010, at 2:36 PM, Mon Ping Wang wrote:

> 
> I don't think so.  They are cases where we want to support copying data from one address space to another through a memcpy intrinsic.  I'll add the isVolatile field to memcpy, memset, and memmove.

Ok, awesome, thanks a lot!

-Chris

> 
> Thanks,
> -- Mon Ping
> 
> On Mar 26, 2010, at 4:44 PM, Chris Lattner wrote:
> 
>> 
>> On Mar 24, 2010, at 4:01 PM, Mon Ping Wang wrote:
>> 
>>> 
>>> This is a patch to add support for address spaces for memcpy, memmove, and memset.  I have changed the signature of these functions to be overloaded based on the pointer type.  I have included both the llvm changes and the llvm-gcc-4.2 changes need to support.  Please let me know if I miss something.  I'll check it in a few days.
>> 
>> Should the source and destination pointers be required to be in the same address space?  I can see it either way (f.e. on x86 a copy from DS -> GS segment address space makes sense), but it would be nice to reduce noisiness in the function name if a copy from one addr space to another isn't useful.
>> 
>> Other than that, this patch looks great to me, but "while you're at it" can you add an "i1 isVolatile" parameter to llvm.memcpy and friends?
>> 
>> -Chris
>