This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] A new meta intrinsic header file for current and future x86 instrinsics.


On Fri, Nov 21, 2008 at 10:01 AM, rajagopal, dwarak
<dwarak.rajagopal@amd.com> wrote:
>> On Fri, Nov 21, 2008 at 9:31 AM, rajagopal, dwarak
>> <dwarak.rajagopal@amd.com> wrote:
>> >>
>> >> On Thu, Nov 20, 2008 at 11:13 AM, Richard Guenther
>> >> <richard.guenther@gmail.com> wrote:
>> >>
>> >> >> So I propose that AMD introduces its own intrinsic file,
> following
>> >> >> immintrin.h example. There *will be* a vendor neutral
> x86intrin.h
>> > (or
>> >> >> whatever the name will be agreed to), and AMDs file will be
>> > included
>> >> >> through this vendor neutral header just in the same way as
> Intels
>> >> >> file.
>> >> >
>> >> > I fail to see a compelling reason to have vendor specific headers
> at
>> >> all.
>> >>
>> >> Me too, but this is the situation we live in (bmmintrin.h vs.
>> >> gmmintin.h) and will always live in. Everybody wants its own
> sandbox
>> >> to play in.
>> >>
>> >> Uros.
>> >
>> > Yes. We are beginning to realize this as well.
>> > I'm planning to create an <x86intrin.h>, which will do the
> following:
>> >
>> > #ifndef _x86INTRIN_H_INCLUDED
>> > #define _x86INTRIN_H_INCLUDED
>> >
>> > #ifdef __MMX__
>> > #include <mmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSE__
>> > #include <xmmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSE2__
>> > #include <emmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSE3__
>> > #include <pmmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSSE3__
>> > #include <tmmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSE4a__
>> > #include <ammintrin.h>
>> > #endif
>> >
>> > #if defined (__SSE4_2__) || defined (__SSE4_1__)
>> > #include <smmintrin.h>
>> > #endif
>> >
>> > #ifdef __SSE5__
>> > #include <bmmintrin.h>
>> > #endif
>> >
>> > #if defined (__AES__) || defined (__PCLMUL__)
>> > #include <wmmintrin.h>
>> > #endif
>> >
>> > #ifdef __AVX__
>> > #include <avxintrin.h>
>> > #endif
>> >
>> > #ifdef __3dNOW__
>> > #include <mm3dnow.h>
>> > #endif
>> >
>> > #endif /* _x86MMINTRIN_H_INCLUDED */
>> >
>> > Is this approach ok?
>>
>> There is no need to include SSE to AVX header files separately.
>>
>> #include <immintrin.h>
>>
>> will just work. I would suggest you include all AMD intrinsic files
> from
>> one file, saying <AMDintrin.h>.  Then x86intrin.h becomes
>>
>> #include <immintrin.h>
>> #include <AMDintrin.h>
>>
>
> Hi H.J,
>
> We have already responded that vendor specific header files are not a
> proposal that is acceptable to AMD.
> We agree with Uros and Richi's views that we should not have vendor
> specific header files. We should instead have architecture extension
> specific header files.
>

It is just a name, nothing more. You can call it <MyISAintrin.h> if you
want.



-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]