• CanadaPlus@lemmy.sdf.org
      0·
      3 days ago

      Okay, now do it by percent of processor (CPU/GPU/whatever) cycles.

      Although, TBF, you can replace it all with C/C++. Or Rust, assume the optimisation has gotten good enough. It’s just that few people are both qualified for and interested in rewriting numerical linear algebra algorithms, and there’s no real reason to if the Fortran works.

  • TimeSquirrel@kbin.melroy.org
    0·
    4 days ago

    Everything is just silicon oxide gates being saturated and drained and turned on and off in various patterns very rapidly in a way that means something to us. That Fortran/C/C++/Assembly depends on that tiny two-MOSFET AND gate in the ALU to do the AND correctly every time.

    • BartyDeCanter@lemmy.sdf.org
      0·
      4 days ago

      Array operations in FORTRAN are much easier for the compiler heavily optimize than it is in c/c++ due to its array model and type system. You can achieve much of the same thing with modern compiler extensions, but it’s difficult and not as portable.

    • tetris11@lemmy.ml
      0·
      4 days ago

      Its just easy to write super-optimised code snippets in without having to break out into assembly.

        • dewritoninja@pawb.social
          0·
          4 days ago

          Not only is it very difficult to write in assembly, the resulting code is not portable. Meaning that if you wrote it on x86 assembly it can’t run on ARM chips without emulation and that takes a significant hit on performance defeating the point

        • SqueakyBeaver@lemmy.blahaj.zone
          0·
          4 days ago

          Yeah, it’s pretty difficult. Think of assembly as just one step above writing 1’s and 0’s, and you’re probably around how difficult it can be

          • recursiveInsurgent@lemm.eeEnglish
            0·
            4 days ago

            I’ve delved into writing assembly only on the level of a student project. I really enjoyed it though. Obviously implementing a python math library would be far more complex but wouldn’t it be worth it for the possible performance gains?

            • Jack Riddle@sh.itjust.works
              0·
              4 days ago

              I don’t think it would be anymore. Modern compilers are really really good at what they do, and often manually optimizing(writing assembly yourself) makes programs slower. So unless you are very good at assembly, I would just trust the compiler.

    • JoshCodes@programming.devEnglish
      0·
      4 days ago

      It provides:

      • a powerful N-dimensional array object
      • sophisticated (broadcasting) functions
      • tools for integrating C/C++ and Fortran code
      • useful linear algebra, Fourier transform, and random number capabilities

      Not according to the repo I sourced from your message.

        • JoshCodes@programming.devEnglish
          0·
          3 days ago

          Yeah look that was the front page of the repo talking about how it has C/C++ and Fortran code, sorry for not reading the docs and finding out that yes they still use C/C++ and Fortran code in the form of OpenBLAS which is a dependency… f2py is just a method of doing the following:

          F2PY facilitates creating/building native Python C/API extension modules that make it possible

          • to call Fortran 77/90/95 external subroutines and >Fortran 90/95 module subroutines as well as C functions;

          • to access Fortran 77 COMMON blocks and Fortran 90/95 module data, including allocatable arrays

          from Python.

          Correct me if I’m wrong here but if you’re implementing an api for one programming language to talk to another then that means you have 2 programm-

          I wake up as a lizard. The meaning of kernels, subroutines and programming languages is already fading. I realise the rock I am lying on is slightly in a shadow and move into the sun. Might eat a bug later…