yaxunl added a comment.
In D75285#1896400 <https://reviews.llvm.org/D75285#1896400>, @hfinkel wrote:
> Unfortunately, we cannot do this kind of thing just because it seems to make
> sense. The language semantics must be exactly satisfied by the IR-level
> semantics. I certainly agree that it would make sense for users to be able to
> mark invariant loads, but this mechanism simply might not be the right one.
>
> One problem here is that, with something like:
>
> char test2(X *x) {
> const char* __restrict p = &(x->b);
> return *p;
> }
>
>
> what happens when the function is inlined? Does the "invariantness" only
> still apply to accesses within the scope of the local restrict pointer? I
> believe that it would not, and that would be a problem because later code
> might legally modify the relevant data.
How about inserting llvm.invariant.end at the end of scope of the variable?
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D75285/new/
https://reviews.llvm.org/D75285
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
