Choosing the Best Fortran Compiler on Linux

When working on a Linux system, the choice of Fortran compiler is crucial, especially considering the standard compliance and performance requirements. This article aims to provide insights into choosing the best Fortran compiler by evaluating key factors such as standard compliance, feature sets, and performance.

Important Factors for Fortran Compilers

Maintaining compliance with the Fortran standards, particularly the 2003 and 2008 standards, is a fundamental requirement. Major compilers like Intel Fortran, IBM, Cray, Absoft, and NAG all produce fast code, but the specific choice can vary based on the context and needs of the project.

Intel Fortran

Intel Fortran stands out for its comprehensive feature set, making it a popular choice among developers. It is particularly preferred for its extensive implementation of features, especially in academic and research environments. Intel also offers the Parallel Studio suite for Linux, which is free for academic use, including the Fortran compiler.

GFortran

GFortran is the best free option and can serve as a reliable choice for projects with budget constraints. Although it may not match the feature set of commercial compilers, it still produces decent performance and is widely used in open-source projects and academic settings.

Feature Implementation in Major Fortran Compilers

The Fortran Wiki provides detailed information on the feature implementations of major Fortran compilers. This resource is invaluable for understanding the capabilities and limitations of different compilers, particularly in relation to the latest Fortran standards.

Performance and Benchmarks

The performance of Fortran compilers can vary significantly, and ongoing competition between leading vendors ensures continual improvement. When evaluating performance, benchmarks such as Polyhedron can provide insight into the capabilities of different compilers. As of the latest benchmarks, Intel Fortran appears to be leading, but the situation is dynamic and can shift based on updates and new developments.

BLAS Libraries and Fortran Compiler Pairings

For most scientific applications, the choice of Fortran compiler is often less critical than the optimization of BLAS (Basic Linear Algebra Subprograms) libraries. Intel Fortran comes with the MKL (Math Kernel Library), which is powerful for many applications. However, OpenBLAS is a slightly faster and more scalable alternative, especially for highly parallel systems. Absoft compilers also work well with the MKL and possibly OpenBLAS, while GFortran can leverage OpenBLAS for optimal performance.

Comparison with C and C Compilers

Today, C and C compilers are generally more advanced in terms of code generation, particularly addressing issues like aliasing that were significant in the past. Improvements in static analysis and processor design have largely resolved these issues, making modern C and C compilers more powerful.

Given these advancements, the choice of a Fortran compiler should not be solely based on speed but on other factors such as feature sets, compatibility, and availability. For many practical applications, the free and open-source GFortran is a solid choice, complemented by optimizations through the use of robust BLAS libraries like OpenBLAS.

In conclusion, while Intel Fortran is a robust and comprehensive choice, GFortran offers a reliable and free alternative. Understanding the needs of your project and evaluating the feature sets and performance of different compilers will help you make an informed decision.