I dont know about OpenCL that seems a specilist case  .   Vector /SIMD
instruction are  crucial  we are approaching 512 bits memcpy  , memcmp ,
bitscan  , just about every program relies on c version of these that use
the 128 and 256 bit versions.  Not to mention doing work on 16 integers at
once is not just an optomization it can be the difference between somehting
being viable and not .  That said as  long as you can trivially link C code
( a worth goal for early adoption) then you can do it in C ( another case
of can you do inline c/asm )  .



> I think if a language system can't support OpenCL (with competitive
> performance), it doesn't deserve to be called a systems programming
> language. The abstraction level at which it injects this kind of
> functionality is a bit irrelevant to me (intern vs library, etc. etc)
>
>
>
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to