Hi Manish,
On 15 Jul 2013, at 10:52, Manish Verma <[email protected]> wrote:
> Your mention of RVCT in your first email caught my attention. The difference
> between the declaration of the intrinsics in the two compilers, is the
> missing "volatile" attribute on the input address argument.
That's a good point, that I hadn't considered. The patch actually doesn't check
the qualifiers at all. Which is probably mostly harmless for the volatile case,
though not ideal. But it also allows storing to a const pointer, which it
probably shouldn't.
It should probably act as if the declarations were:
__builtin_arm_ldrex(const volatile T *);
__builtin_arm_strex(T, volatile T *);
Thanks for spotting that, I'll fix it before sending the patches again.
I think adding an implicit cast on the pointer expressions to the appropriately
qualified type should be sufficient.
Tim.
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
