The Modern Flang "f18" Compiler Is The Most Exciting Fortran Compiler Of Recent Times

[phoronix]

Phoronix: The Modern Flang "f18" Compiler Is The Most Exciting Fortran Compiler Of Recent Times

While merging of the Flang "f18" Fortran compiler into the LLVM source tree was delayed in January, this is still looking like the most exciting Fortran open-source compiler in development...

http://www.phoronix.com/scan.php?page=news_item&px=Flang-f18-Looking-Great-2020

[Setif]

The most exciting news is that it will not be an alternative to gfortran until the end 2021.

[wizard69]

The most exciting news is that it will not be an alternative to gfortran until the end 2021.

Why would that be a positive? The llvm suite is the best thing to happen to the GCC world. I honestly believe that GCC would not be where it is today without the push from LLVM/CLang. GFortran will benefit in the same way.

there is as a time when compiler development wasn’t even news worthy. Now we have a well developed GCC and CLang that meet standards. Both perform very well and have come a very long way in making the developers life easier.

[cesarcafe]

The first LLVM release to have f18 included will be wonderful. And the second one, with things more polished after such a big merging, will be a dream come true. I really look forward to see this happen!!

[Anarchy]

Should have added a tl,dr. :/

This new flang compiler project is mostly run by nvidia... I assume because the pgi compiler never actually became good enough. So, now, I guess they're trying to make something better that will likely interface well with other nvidia software. It might take off, but you never know with these projects.

One of the slides in the presentation is also a bit misleading, they're talking about "performance" but they don't include comparisons with Intel's fortran compiler, which currently is *the* industry standard. Anyway, it will be interesting to see how this shapes up in the future. Usually when it comes to gfortran, it's always the lack of new features that is the problem.

[AKing]

Should have added a tl,dr. :/
Usually when it comes to gfortran, it's always the lack of new features that is the problem.

GNU Fortran (along OpenCoarrays) actually has done a really good job of keeping up with the standard. Full (or near full) support of 2003 + a very big part of Fortran 2008 and 2018. Much of my lab's codebase is entirely modern Fortran 2008 complaint. I am surprised to see that the same code can be compiled with almost no change via both Intel 2018 and GNU 7 compilers. Of course, they have added a lot more beyond GCC 7. But I confess that bugs are still abundant in GFortran, as compared to Intel ifort.

[Anarchy]

GNU Fortran (along OpenCoarrays) actually has done a really good job of keeping up with the standard. Full (or near full) support of 2003 + a very big part of Fortran 2008 and 2018. Much of my lab's codebase is entirely modern Fortran 2008 complaint. I am surprised to see that the same code can be compiled with almost no change via both Intel 2018 and GNU 7 compilers. Of course, they have added a lot more beyond GCC 7. But I confess that bugs are still abundant in GFortran, as compared to Intel ifort.

That's interesting. I've hit a number of bugs in ifort over the years, which were all confirmed as bugs and fixed by Intel, but I've never actually hit a gfortran bug. Usually with me, it was that I wanted to use some new fortran feature that wasn't supported by the compiler. Although, I have to say that the last time I used fortran was two years ago, I've since switched jobs, maybe in the meantime gfortran became even better.

Anyway, it would actually be really interesting to see what they're gonna achieve with this new compiler. I remember trying to use pgi compilers in the past, but they were really not that great, much of my code did not work at all with them.