On 13 August 2014 17:05, Renato Golin <[email protected]> wrote:
> On 12 August 2014 15:36, Yi Kong <[email protected]> wrote:
>> ARM supports instruction prefetch which the generic prefetch intrinsic
>> is not able to describe. AArch64 has even more granular instructions.
>
> Hi Yi,
>
> I assume you'll map ACLE __p* into variations of Intrinsic::prefetch
> in the same way.
>
> If that's so, why do you need to add __builtin_arm_prefetch to either
> target? Why not add __pldx as the generic one and get __pld/__pli as
> wrappers around that? Is there already usage of such a builtin
> elsewhere?
>
__pldx is not an obvious name for generating instruction prefetch as
well as data prefetching, especially when trying to use the intrinsic
directly instead of through the arm_acle header. Also, it is a
Clang/GCC convention to have target-specific builtin intrinsic name to
start with __builtin_{target name}.
-Yi
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
