We received a lot of positive feedback on our LFortran announcement. Most generally like the idea and tool, and expressed interest to hear a bit more on why we think Fortran is a superior language in its domain and why it makes sense to use Fortran for new projects.
Why Fortran?
Fortran was designed from the ground up to naturally and simply translate mathematics to code that compiles and runs at maximum speed. And being specifically targeted for such fundamentally computational tasks, it contains a broad range of key functionality within the language itself, standard across all platforms, with no need for external libraries that may or may not be well optimized or maintained, at present or down the road.
Some highlights:
- Multidimensional arrays which can be allocated and indexed as the math/science dictates (not restricted to start at 0 or 1) and can be sliced as desired (as, e.g., in MATLAB);
- Operators which operate naturally upon the aforementioned arrays/matrices, as they do scalars;
- Complex numbers;
- Special functions;
- Structures and pointers for more general data representation.
Because the essentials are contained in the language itself, it is simple to read and write, without need of choosing from among or deciphering a proliferation of external classes to do the same thing. And because the essentials are self-contained, compilers can provide detailed compile-time (e.g., argument mismatch) and run-time (e.g., memory access) checks, as well as highly optimized executables, directly from natural, readable code without need of extensive optimization heroics by the developer.
If you are interested in learning more, please see our webpage at fortran90.org with recommended practices for writing code, side by side comparison with Python/NumPy, links to other online Fortran resources and books, and an FAQ.
The above distinguishing aspects, among many others, have made Fortran the language of choice in high-performance computing for decades. Indeed, by virtue of the simplicity of expressing the desired mathematics and corresponding ease of maintenance by application scientists/engineers (i.e., by the users themselves), the majority of HPC application codes in widest use today are written mostly or completely in Fortran (e.g., 7 out of the top 10 HPC codes from this 2017 NVIDIA survey are written in Fortran). And yet, new projects are being started in Fortran less and less, despite its manifest advantages.
The reasons for this, we believe, are clear, ranging from lack of freely available and modern tooling to lack of exposure in the education pipeline, as elaborated in our previous blog post.
The goal of this project is to address these needs and, in so doing, bring the significant advantages of Fortran to the full range of computational science and engineering applications which stand to benefit.
Trackbacks: