[LLVMdev] GSoC project questions.

Wed Apr 24 11:58:28 PDT 2013

----- Original Message -----
> From: "Matthieu Brucher" <matthieu.brucher at gmail.com>
> To: "Alex L" <arphaman at gmail.com>
> Cc: "Hal Finkel" <hfinkel at anl.gov>, "LLVM Developers Mailing List" <llvmdev at cs.uiuc.edu>
> Sent: Wednesday, April 24, 2013 10:12:26 AM
> Subject: Re: [LLVMdev] GSoC project questions.
> 
> 
> Hi Alex,
> 
> 
> You may want to check the later versions of the standard also to
> check if some of your proposal is not deprecated there. The most
> important thing in the Fortran standard is the array support and it
> is a big task by itself. In my opinion, the I/O part is not the most
> important thing yet it is an enormous task. Usually, the Fortran
> routines are used externally in other computation and I/O is done
> through other and more efficient layers.

We should double-check with Chris Lattner about license compatibility, but regarding basic Fortran I/O, we can probably take some base code from libf2c. That should give us the correct Fortran 77 I/O semantics without too much pain (http://www.netlib.org/f2c/). http://www.netlib.org/f2c/src/notice -- it seems BSD-like. I'm suggesting this because my feeling is that most of the work in I/O is really in the runtime, not in the semantic-analysis/codegen pieces.

Also, I agree that we can put off I/O support for some time (perhaps even until after the summer); a large part of modern Fortran use, although certainly not all of it, is in writing kernels that are called from C/C++ programs. In this use case, Fortran I/O is specifically avoided.

 -Hal

> 
> 
> I hope your proposal will be accepted and I'm looking forward to
> seeing the result of your work.
> 
> 
> Cheers,
> 
> 
> Matthieu
> 
> 
> 
> 2013/4/24 Alex L < arphaman at gmail.com >
> 
> 
> 
> 
> Hi again.
> 
> 
> I've though more about the full standard being too ambitious, and now
> I fully agree with that. So I had a go at choosing the useful subset
> of Fortran on which I would like to work on - my update proposal can
> be viewed at https://gist.github.com/hyp/5434845 .
> 
> Focusing on less should also improve my productivity and lead to a
> better summer of code project by drastically reducing the pressure
> :)
> 
> 
> 
> 
> 
> 
> 2013/4/22 Alex L < arphaman at gmail.com >
> 
> 
> 
> 
> 
> Thanks for the responses!
> 
> 
> 
> 
> I might be wrong (I'd love to be wrong here!) but I think this is a
> bit too ambitious
> You might be right about it being too ambitious, it would certainly
> be wise to aim for less. A more polished, but less broad product is
> definitely better than a buggy standard compliant one, and I think
> that it would go better with the general philosophy of GSoC. I will
> take this fact into consideration for the revised proposal.
> 
> 
> 
> 
> Also, in your proposal, it looks as though you're not writing tests
> until the end; tests need to be written as soon as the features are
> testable.
> 
> That's kind of true, but I actually planned on writing tests during
> feature development, it's just that these tests wouldn't cover
> everything and should be expanded and tested fully towards the end
> of GSoC. I should definitely mention that in my plan.
> 
> 
> As for the tooling support, I guess that it wouldn't hurt to take the
> fact that tools should be integrated with flang in the future into
> account during the development. I will take a look on how clang and
> various tools for it operate.
> 
> 
> 
> 
> 
> 
> 2013/4/22 Hal Finkel < hfinkel at anl.gov >
> 
> 
> 
> ----- Original Message -----
> > From: "Dmitri Gribenko" < gribozavr at gmail.com >
> > To: "Alex L" < arphaman at gmail.com >
> > Cc: "LLVM Developers Mailing List" < llvmdev at cs.uiuc.edu >
> > Sent: Monday, April 22, 2013 9:00:51 AM
> > Subject: Re: [LLVMdev] GSoC project questions.
> > 
> 
> > On Mon, Apr 22, 2013 at 4:30 PM, Alex L < arphaman at gmail.com >
> > wrote:
> > > Thanks for your support.
> > > I wrote my draft GSoC proposal, it can be viewed here -
> > > https://gist.github.com/hyp/5434845
> > > Please read it and tell me what you think. Any criticisms or
> > > suggestions are
> > > welcome!
> > 
> > Hello Alex,
> > 
> > > I plan to make flang a fully featured frontend which fully
> > > supports
> > > Fortran 77
> > 
> > I might be wrong (I'd love to be wrong here!) but I think this is a
> > bit too ambitious. Given that you have to write tests *as you
> > implement the features*, handle corner cases (and one has to
> > actually
> > discover corner cases from the standard, if one is not familiar
> > with
> > Fortran) and produce diagnostics for invalid code, I don't think it
> > is
> > possible to implement all of the frontend + codegen within 12
> > weeks.
> 
> Alex, I agree with Dmitri, I think that the proposal sounds a little
> too ambitious. Maybe everything will go smoothly and you'll be able
> to accomplish everything that you'd like, but we should plan for
> less. I favor identifying a useful subset (such as that needed for
> BLAS) and focusing on doing a really good job on that subset rather
> than trying to get full feature coverage. In my experience, coding
> tests and high-quality warning/error messages takes more time than
> coding the feature itself. Also, in your proposal, it looks as
> though you're not writing tests until the end; tests need to be
> written as soon as the features are testable.
> 
> Regarding motivation, I would also like to bring the same kind of
> "tooling" (analysis and transformation) support available in Clang
> to Fortran. Not that this specifically impacts your project, but I
> think is an important part of the longer-term motivation for
> developing an LLVM Fortran frontend.
> 
> I am glad that you're excited about working on this project!
> 
> -Hal
> 
> 
> > 
> > Dmitri
> > 
> > --
> > main(i,j){for(i=2;;i++){for(j=2;j<i;j++){if(!(i%j)){j=0;break;}}if
> > (j){printf("%d\n",i);}}} /*Dmitri Gribenko < gribozavr at gmail.com
> > >*/
> 
> 
> > _______________________________________________
> > LLVM Developers mailing list
> > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> > 
> 
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 
> 
> 
> 
> 
> --
> Information System Engineer, Ph.D.
> Blog: http://matt.eifelle.com
> LinkedIn: http://www.linkedin.com/in/matthieubrucher
> Music band: http://liliejay.com/
> 