Announcement

Collapse
No announcement yet.

NVIDIA Has Been Working On A New Fortran "f18" Compiler It Wants To Contribute To LLVM

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    FORTRAN is still being used, but as far as I know that's limited to legacy code mostly. That code is still in production and works fairly well because the generated machine code is just right, although you probably wouldn't want to use FORTRAN for anything new. It isn't even nearly as common as it used to be, most people in formerly FORTRAN-loving fields are using Matlab, R, Python etc. to get things done, Even if there's still some FORTRAN under the hood in those environments, very few people deal with it or are capable of writing high-performance code.

    Especially when it comes to highly-optimized performance-sensitive code, I don't see much point in changing compilers now. By the time you delve into FORTRAN and get the toolchain and your code in a good shape, you might as well write target-specific machine code for critical stuff and call it from just about any high-level language.
    Last edited by edgmnt; 03-02-2019, 05:10 PM.

    Comment


    • #12
      Originally posted by hotaru View Post

      I'll be surprised if Nvidia ever contributes anything of real value (open source GPU drivers, for example).
      You never know! If AMD can continue with their recovery and Intel introduces a GPU with real compute ability NVIDIA could start to feel the heat.

      writing a new Fortran compiler is like designing a new 56k modem. it may have some niche uses, but to most people it's completely worthless.
      I’m not even sure why I’m reading this thread. Fortran has always been a bit of a joke from my perspective. It is sort of like COBOL in that regard.

      Comment


      • #13
        Originally posted by edgmnt View Post
        FORTRAN is still being used, but as far as I know that's limited to legacy code mostly. That code is still in production and works fairly well because the generated machine code is just right, although you probably wouldn't want to use FORTRAN for anything new. It isn't even nearly as common as it used to be, most people in formerly FORTRAN-loving fields are using Matlab, R, Python etc. to get things done, Even if there's still some FORTRAN under the hood in those environments, very few people deal with it or are capable of writing high-performance code.

        Especially when it comes to highly-optimized performance-sensitive code, I don't see much point in changing compilers now. By the time you delve into FORTRAN and get the toolchain and your code in a good shape, you might as well write target-specific machine code for critical stuff and call it from just about any high-level language.
        Fortran is more or less the language of choice of engineers and scientists, it's also widely used in the aerospace industry and similar industry branches where engineers and scientists write code that has to be fast and dependable; that's because Fortran is easy to learn, very robust, and comes with extremely efficient multidimensional arrays that are part of the language in addition to a large math related standard library. Additionally, the well established Fortran compilers are very good; the Intel Fortran Compiler, btw, is probably the best compiler I've ever used for any language.

        Recently, there's been a slight shift in how code is developed, it used to be that people wrote everything in one language. Or tried to mix Fortran with C/C++ or do something similar, but nowadays it's common to develop the computational code in Fortran and then just write thin wrappers in Python on top of it. If you use fortran this is basically as easy it can possibly be, you just need to issue a few commands and you're done. C & C++ are also a common choice, but to be frank, if you're not very comfortable with them you can easily end up with an extremely buggy code, and C++ was not a choice at all for many years because the array implementations sucked and sometimes due to leaky abstractions to fix a weird problem you had to go and read the underlying code of whichever library used. Also, the C++ compilers were a nightmare.

        One way or another, Fortran isn't going anywhere, it's a niche, but it's been like that for something like two-three decades, which is perfectly fine, it's a language for heavy computations, it's not a language, in which you want to write a performant regexp library.
        Last edited by Anarchy; 03-02-2019, 08:12 PM.

        Comment


        • #14
          For heaven's sake, to those commenters who don't know anything about Fortran: It's been 3 decades since FORTRAN was replaced by Fortran. NO NEED FOR CAPS LOCK.

          Comment


          • #15
            Fortran is still used heavily, in High-Performance Computing circle. Based on the past 5 years of my work at the #1 computational science research institute of the world, I can tell you that there are many researchers and industries that use and develop software in Fortran, and there are prominent research groups that are using the latest features of Fortran 2018 and 2008 to develop such software. Of course, it is true that the field does not exclusively belong to Fortran anymore. After all, the landscape of programming languages is quite different now from what it was in the 1980s. Also, I can tell you that Intel compiler is getting a lot of attention from Fortranners, because of their efforts to implement the latest Fortran standard features, already covering the entire 2008, and even parts of Fortran 2018, which is impressive, in my opinion.

            Comment

            Working...
            X