[PATCH] D18164: [tsan] Do not instrument reads/writes to instruction profile counters.
Xinliang David Li via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 15 11:16:42 PDT 2016
On Tue, Mar 15, 2016 at 10:43 AM, Anna Zaks <zaks.anna at gmail.com> wrote:
> zaks.anna added a comment.
> dvyukov's point is that there are no "benign races" based on LLVM
> semantics. People might argue that their race is benign because it will
> result in atomic reads/writes on the architectures they care about.
It is not about this assumption, but more about racy counter updates not
being a problem practically (other than the precision of the profile data
might be affected slightly in practice).
> However, in that case, they should annotate their loads and stores as
> atomic explicitly, which would lower to no-op on the architectures they
> care about but will also be modeled correctly at the LLVM level. (TSan's
> pass reasons about LLVM semantics.) This is why I suggested to use
> "Monotonic". (http://llvm.org/docs/Atomics.html#id10)
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits