[cfe-dev] clang and gcc X86 built-in functions

[Alexandre Colucci]

> 
> On Sep 16, 2010, at 10:50 AM, Sebastian Redl wrote:
> 
>>> 
>>> #include <stdio.h>
>>> 
>>> int main (int argc, const char * argv[])
>>> {
>>>     typedef short _SSE_VecS16 __attribute__ ((__vector_size__ (16)));
>>>     
>>>     register _SSE_VecS16 b;
>>>     register _SSE_VecS16 c;
>>>     __builtin_ia32_paddw128(b, c);
>>>     
>>>     return 0;
>>> }
>> 
>> IIRC, Clang intentionally doesn't have any built-ins for operations which are naturally modeled by operators. Either use the official SSE intrinsics, or use operators. Each built-in has a maintenance overhead, and such redundant ones aren't worth it, when b + c (or b += c?) does the same thing.
>> 
> 
> More specifically, please use the <xmmintrin.h> and related headers.  Those are the public, documented interfaces to the X86 vector instructions.  Using generic vectors is also a great option if you only care about targeting clang, not other compilers like GCC, ICC, MSVC etc.
> 
> -Chris

Thanks for the info. We switched to <xmmintrin.h> and it now works as expected.

Alexandre
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20100929/e7f95146/attachment.html>