flang and IFX

[AndrewC]

I was wondering if someone could shed light on the relationship between the two LLVM based Fortran compilers flang and IFX. Is there cross-pollination or are they completely separate development branches?

[Ron_Green · Accepted solution]

Completely separate.  We use our own Intel Fortran Front End.  Flang-new is an open source front end with vendor funded development assistance.

The only similarity is that the 2 use the LLVM framework.  

[AndrewC]

Hi Ron,

Thanks for the update. That is interesting.  Seems like a lot of effort when IFX is free to download and the Fortran community is small.

Andrew

[Jacob_Williams]

Don’t forget there are other processors than Intel. 

[Steve_Lionel]

There are some users who want an all-open-source option. Also, the work on the current flang started before Intel announced it was moving to LLVM. My understanding is that the flang project is being financed in part by a US government contractor.

[Ron_Green]

It is my opinion, and this may or may not reflect the that of my employer Intel, is that we need a healthy and robust set of Fortran compilers from both open source and vendors.  I think the biggest threat to our community and Intel Fortran is the urban myth that "Fortran is a dead language".   We are in fact in the top 10 programming languages and growing in popularity as cataloged by the Tiobe Index.   But this only happens if we, as a overall Fortran community, have reliable and efficient compilers for every conceivable architecture and OS.   And on those targets, having more that 1 compiler to choose from.  If you follow this Forum or the fortran-lang Discourse group, you will see we (the community) often use multiple compilers to test conformance of complex declarations, statements, and expressions. If code from one compiler crashes the compiler, or hits an odd runtime error, what do the other compilers do with this code?  Often one or another compiler may give a diagnostic that helps isolate and explain an issue, either with the compiler itself or the program in question.  

Open source compilers are a great resource for the community.  Gfortran, LFortran, and now Flang (flang-new) and others provide support for a wide variety of OSes and platforms.  Vendors provide compilers that are highly tuned for their hardware and OS offerings, but lack the coverage over competing hardware.  And as mentioned, there are non-Intel architecture processors gaining in popularity.  And vendor accelerators that need support from their vendor compiler to produce efficient code that justify the additional cost in power, cooling, and cost., and most importantly the high cost in code modifications and maintenance.  

So it's important that efforts like flang succeed.  As a community we need choice in open source compilers.  Creating a MODERN Fortran front end is a complex effort, on the order of 7-12 years with about a dozen or more full time dedicated developers (the complexity of the code makes it intractable for the casual, part-time hobbyists).  It's non-trivial.  Trust us on that. Several vendors are counting on using flang as a basis for their own highly tuned compiler to support their platforms.  And many large organizations heavily invested in Fortran need choice in compilers along with support for modern Fortran language features. This helps with funding to keep developers paid and dedicated and to keep the development on track and progressing to reach maturity.  It's a fascinating effort, and one that I hope succeeds in the long run.  Again, bringing flang to maturity and stability is and will continue to be a multi-year effort.  Anyone who has worked on any modern Fortran compiler knows the effort and years of dedication needed to bring a compiler to maturity.