[llvm-dev] [flang-dev] About OpenMP dialect in MLIR

[Johannes Doerfert via llvm-dev]

On 02/19, Vinay Madhusudan via flang-dev wrote:
> I think it is best to have the design / implementation for few more
> constructs to get more insights. I believe that the best course at this
> point would be to wait for the results and act accordingly.

We do have an overall design and we are starting the implementation
right now. As part of this implementation we'll add more constructs and
we'll gain insights. Based on these insights we'll continuously adjust
our design and implementation.

I'm unsure what else you mean by "waiting for results and act
accordingly". Could you elaborate what exactly you want us to do?


> Thanks for the detailed and insightful discussion,
> Vinay
> 
> On Tue, Feb 18, 2020 at 11:12 PM Johannes Doerfert <jdoerfert at anl.gov>
> wrote:
> 
> > On 02/18, Vinay Madhusudan via flang-dev wrote:
> > > Given that there are unconcluded things in this thread about OpenMP
> > > representation for basic constructs (including target constructs) in MLIR
> > > and OpenMPIRBuilder being the high level *common* interface for Clang AST
> > > and optimized MLIR IR, I believe that it would be a good idea to wait for
> > > things until there is more clarity on OpenMP in MLIR.
> >
> > I agree, there are open questions. However, most of them need empirical
> > data to be answered properly. We cannot accurately predict what may or
> > may not work wrt. optimizations that may or may not exist in the future.
> >
> > What I try to say is that it seems to me the path forward described by
> > Kiran will bring us closer to where we want to be, at least as far as we
> > can describe that point at this time. What we should aim for is an
> > approach that generates useful results (=code + insight) regardless of
> > potential directions changes mid-way. We obviously have to reevaluate
> > choices as we go, meaning the plan is not set in stone, but that is
> > always the case anyway.
> >
> > Please let me know if
> >  - you disagree with this,
> >  - you think the plan presented by Kiran will prevent us from
> >    reevaluating choices along the way and changing course accordingly,
> >  - you think there is a realistic chance the path will cause a
> >    non-trivial amount of wasted work.
> 
> Thanks,
> >   Johannes
> >

> _______________________________________________
> flang-dev mailing list
> flang-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/flang-dev


-- 

Johannes Doerfert
Researcher

Argonne National Laboratory
Lemont, IL 60439, USA

jdoerfert at anl.gov
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200219/55879061/attachment.sig>