[cfe-commits] please review: adding module-level metadata for ThreadSanitizer

[Kostya Serebryany]

"Encoding Compile Flags into the IR"?
Yes, that's the same problem.
Great! I can wait.

--kcc

On Wed, Aug 22, 2012 at 9:55 PM, Eric Christopher <echristo at apple.com>wrote:

>
> On Aug 22, 2012, at 12:20 AM, Kostya Serebryany <kcc at google.com> wrote:
>
> > Hello,
> >
> > I'd like to resurrect an old discussion: how to selectively disable
> ThreadSanitizer (or AddressSanitizer, etc) instrumentation
> > for particular functions.
> > For AddressSanitizer we currently use a function attribute
> "AddressSafety" (if it is not set, instrumentation does not happen),
> > but Chris doesn't like this idea (attributes are a scarce resource) and
> he suggested to use module-level metadata instead of an attribute.
> >
> > The attached clang patch adds such metadata node for all functions when
> -fthread-sanitizer is given.
> > If that is what we want, I'll gradually change AddressSanitizer to use
> similar approach and will kill the "AddressSafety" attribute.
> >
> > BTW, the code looks like this:
> >   llvm::Value *Fn[] = { F }; ...
> >   llvm::MDNode::get(VMContext, Fn);
> > Do we want this code to look like
> >   llvm::MDNode::get(VMContext, F);
> > (i.e. to add more llvm::MDNode::get() methods)?
>
> Bill has been working on something similar for command line options and
> LTO. There's still been some chatting about it, but Bill's planning on
> updating the list in the next few days with what's going on if waiting a
> bit is ok?
>
> -eric
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20120823/94410d79/attachment.html>