LLVM's Fortran Compiler "Flang" Makes Significant Progress But Not Yet Production Ready

[phoronix]

Phoronix: LLVM's Fortran Compiler "Flang" Makes Significant Progress But Not Yet Production Ready

Being worked on actively by a number of stakeholders in recent years has been Flang as a Fortran language front-end to the LLVM compiler stack. While not yet ready for general use, Flang has been making strides as well as showing some promising performance results...

https://www.phoronix.com/news/LLVM-Flang-Early-2023

[kylew77]

Fortran is a cool language. Been around longer than C and is comparable or faster in speed. Fortran is so old it predates the terminal and was used on punch cards! I always wanted to take a class on it and one of the universities I attended offered a course in it but the professor in charge of what classes ran would never run it because he didn't consider it relevant. Would offer Matlab instead of Fortran for crying out loud!

[stormcrow]

Fortran is a cool language. Been around longer than C and is comparable or faster in speed. Fortran is so old it predates the terminal and was used on punch cards! I always wanted to take a class on it and one of the universities I attended offered a course in it but the professor in charge of what classes ran would never run it because he didn't consider it relevant. Would offer Matlab instead of Fortran for crying out loud!

Apparently that professor wasn't an experimentalist.

I had to learn Fortran in order to operate in the experimental physics world. It's a very easy language to learn. I haven't used it since the '77 days, but I understand the modern revisions have done away with the formatting requirements that are a hold over from the punch card days.

[Ironmask]

I remember a Microsoft story where MS made a Fortran compiler, and an employee asked one of the marketing guys how many they sold so far. He said 2. The employee asked "oh cool, 2,000?" "No, 2."

[stormcrow]

I remember a Microsoft story where MS made a Fortran compiler, and an employee asked one of the marketing guys how many they sold so far. He said 2. The employee asked "oh cool, 2,000?" "No, 2."

I know one of the two licenses sold. I used it. That said, it's probably apocraphal. There's probably at least 3 licenses . Back then Fortran on an MSDOS PC wasn't really something many experimental scientists did. PCs were still "toys" and even worse, they ran a toy OS called Microsoft DOS which couldn't even fully use the hardware even on a 286 let alone a 386 or 486... "Doesn't even have proper pipes!" as one big iron user put it in '93. There were a rare few PCs that had some flavor of UNIX, but they were expensive (license costs for the Unices were stupidly expensive even on a PC) and weak sauce hardware in comparison to RISC workstations and compute systems of the time. Computational physics wasn't yet a separate specialization field. It was still a part of traditional experimental physics.

Most Fortran was Fortran 77 on RISC based UNIX systems with a few VMS and IBM zealots sprinkled in for seasoning. Microsoft was good enough for secretaries, but no "real" scientist used it except to write a proposal for "real computers".

As an aside, I wiped a condescending smirk off the face of a VMS zealot around '95 when I managed to get a cobbled together 486DX PC that sometimes barely functioned (hard drive spindle had a tendency to stick) to get better sample rates than an embedded MicroVAX that cost 10x as much with a much simpler program written in C than the machine code they had to use on the VAX. It was really just a minor thing technically, but it's less the achievement than proving to a condescending arrogant jerk that his assumptions were badly antiquated and his job was in real danger even from undergraduates. That's the time period when big expensive experiments weren't being funded by the NSF in the aftermath of the SSC demise and contributed to the beginning of the decline of the traditional Big Iron dominance in the computing industry and scientific computing.

[kvuj]

Executables can be created • Use the `flang-experimental-exec` flag*

Does it mean it can finally create an executable without gfortran? If so, that's sick!

[CommunityMember]

I had to learn Fortran in order to operate in the experimental physics world.

A lot of the classic HEP algorithms were originally written in fortran, and there was no specific reason to want to rewrite them in another language which would almost certainly run slower (and no HEP lab wanted their code to run slower), and one would have to do the validation of results of those libraries all over again. Of course, times have changed, but CERNLIB had a very long useful lifetime (replaced by ROOT).

formatting requirements that are a hold over from the punch card days.

Those were typically 5081 forms (the form number of the classic IBM 80 column punch card), and those formatting standards (which recommended sequence numbers in col 72-80) made it possible to have a card sorter put your deck back in order after you dropped it on the floor.

[archkde]

A lot of the classic HEP algorithms were originally written in fortran, and there was no specific reason to want to rewrite them in another language which would almost certainly run slower (and no HEP lab wanted their code to run slower), and one would have to do the validation of results of those libraries all over again. Of course, times have changed, but CERNLIB had a very long useful lifetime (replaced by ROOT).

Isn't there still a large number of linear algebra software written in Fortran out there? Someone said once that Numpy is basically a Python interface to Fortran, and I think this is still mostly true.

[toguro123]

Fortran 2003 got me through grad school. Y'all may start feeling old now...

[jrdoane]

At the university I went to, I believe a friend of mine who was a meteorology major said that they still (back in 2010-ish,) needed to learn Fortran to do weather modeling and whatnot.