Many open-source Fortran codes are listed by topic at https://t.co/JdtNnNmn1h and https://t.co/bZmof1LhWr

https://t.co/17NyFm7v6u To simplify compiler bug reports, Damian Rouson suggests that 1. All procedure definitions go in submodules 2. Each submodule’s immediate ancestor is a module (i.e., there’s no intervening submodule) 3. Each file contains only one module or one submodule.

Had an interesting discussion with a colleague about the merits of #Fortran and #Python for an undergrad class (ME not CS) on numerical linear algebra. Having just migrated my personal webpage, I figured I'll start a technical blog and write about it. 👉

New post about 𝗙𝗼𝗿𝘁𝗿𝗮𝗻. We'll explore how to go from a naïve implementation of the Jacobi method taking a whole minute to solve the 2D Poisson (~250 000 unknowns) to a multithreaded one solving the same problem in less than 3 seconds. 👉

Related discussion:

Current support by Intel Fortran of the Fortran 2023 standard and OpenMP is summarized in a table at https://t.co/YC72lJdmLa .

https://t.co/13OxdRD9wV Damian Rouson: "default(none) in do concurrent is good code hygiene – similarly to implicit none. Using default(none) in a do concurrent construct obligates the programmer to specify the locality of variables, e.g., local, local_init, or reduce."

https://t.co/pxIHUuKWdJ "Fortran 2023: When, How, and Why We Use it", by Damian Rouson

https://t.co/wYwvonj5hn Intel Fortran compiler 2025.2.0 released. They suggest also installing Intel Distribution for GDB, Intel MPI Library and oneMKL.

An auto-generated, unofficial introduction to the Fortran Standard Library is at https://t.co/11OqjyGbYI. DeepWiki can summarize any GitHub project.

For a zero-size 1-D array x, lbound(x) == 1 and ubound(x) == 0, so it is still true that size(x) == ubound(x) - lbound(x) + 1 For zero-size logical array x, all(x) .eqv. .true. and any(x) .eqv. .false. so the property all(x) .eqv. (.not. any(.not. x)) still holds

https://t.co/twLKuN5TVO David Pfister created a tool to display the module dependencies of a Fortran project.

https://t.co/dTiBxDkDbl By prohibiting the overriding of a parent type’s type-bound procedures and using PURE procedures with DO CONCURRENT, the authors matched the performance of OpenMP. The deep learning library used is

Fortran features added and removed, from the Original IBM 704 FORTRAN through Fortran 2018, are listed at

Using QuickWin with Intel Fortran, you can make Fortran GUI programs for Windows. Another choice is Silverfrost FTN95’s native Clearwin+ library.

Fortran compares strings of unequal LEN by padding the shorter string with blanks at the end. Therefore you can write if (s == "yes") if len(s) > 3 instead of if (trim(s) == "yes")

At https://t.co/jpTUXM36kB is a tool by Ubaid Shaikh that shows that Abstract Syntax Tree and Abstract Semantic Representation of a Fortran code, using LFortran.

DO CONCURRENT communicates that an operation should be independent of order, but when there is no parallelization method the compiler must pick some order to follow. Marshall Ward ran some tests and suggests outer to inner loops.

OpenMP directives can be faster than DO CONCURRENT, since compiler writers have put more effort in implementing the former

Slides for an Intermediate Modern Fortran course covering the facilities for abstraction and polymorphism provided by classes and interfaces, and the facilities for interoperability with C are at

The UK Met Office is giving a course "Introduction to Modern Fortran", with slides, including exercises, at