[llvm-dev] RFC: Representing unions in TBAA
Hal Finkel via llvm-dev
llvm-dev at lists.llvm.org
Thu Mar 9 11:27:21 PST 2017
On 03/09/2017 12:58 PM, Flamedoge wrote:
> As far as my limited understanding goes, this should be a bugfix
> (correctness) and it should trump backward compatibility requirement.
> It makes no sense to me to continue backward compatible incorrectness.
The TBAA scheme is internally self-consistent and there are other
(non-C/C++) frontends that are using it in a completely well-defined
way. The bug, as such, is in the mapping between C/C++ and the TBAA
metadata. That, indeed, we'll just change once we have something to
target that can correctly represent the required semantics.
-Hal
>
> Kevin
>
> On Thu, Mar 9, 2017 at 10:21 AM, Steven Perron via llvm-dev
> <llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>> wrote:
>
> I could make something like Daniel's suggestion work. The main
> question I still have is how we tell the difference between the
> "old-style" DAG and the "new-style" DAG? I don't know if there is
> some standard way of doing this. Do we just base it on the
> version of llvm-ir?
> Would there be a need to maintain different code for both type of
> DAG? If we change clang to generate the new-style DAG, things
> will be fine for C/C++. However, will this force other components
> that generate llvm-ir to change?
> In general, what kind of backwards compatibility do we need to keep?
> Later,
> Steven Perron
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>
>
--
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170309/b81286c7/attachment-0001.html>
More information about the llvm-dev
mailing list