High-Performance Design Patterns for Modern Fortran

Abstract

This paper presents ideas for using coordinate-free numerics in modern Fortran to achieve code flexibility in the partial differential equation (PDE) domain. We also show how Fortran, over the last few decades, has changed to become a language well-suited for state-of-the-art software development. Fortran’s new coarray distributed data structure, the language’s class mechanism, and its side-effect-free, pure procedure capability provide the scaffolding on which we implement HPC software. These features empower compilers to organize parallel computations with efficient communication. We present some programming patterns that support asynchronous evaluation of expressions comprised of parallel operations on distributed data. We implemented these patterns using coarrays and the message passing interface (MPI). We compared the codes’ complexity and performance. The MPI code is much more complex and depends on external libraries. The MPI code on Cray hardware using the Cray compiler is 1.5–2 times faster than the coarray code on the same hardware. The Intel compiler implements coarrays atop Intel’s MPI library with the result apparently being 2–2.5 times slower than manually coded MPI despite exhibiting nearly linear scaling efficiency. As compilers mature and further improvements to coarrays comes in Fortran 2015, we expect this performance gap to narrow.

The Scalability-Efficiency/Maintainability-Portability Trade-Off in Simulation Software Engineering: Examples and a Preliminary Systematic Literature Review

“…Paper [37] investigates how a DSL can provide better maintainability while keeping good efficiency. Paper [38] describes design patterns for FORTRAN for modern, maintainable software which is still efficient. Paper [39] reports on the experience of using existing frameworks especially for infrastructure and data management to improve extensibility while keeping high efficiency.…”

“…Only one paper was published as a technical report and, hence, definitely non-refereed. [33], [34], [35], [36], [37], [27], [40], [30], [41], [31], [43], [32], [44] Empirical study without intervention 11 [18], [22], [24], [26], [15], [29], [19], [42], [20], [17], [21] Empirical study with intervention 5 [23], [38], [39], [25], [28] Opinion/philosophical paper 4 [13], [14], [2], [16] We also classified the included article in different types. We show the results in Tab.…”

“…Furthermore, we identified four different classes of contributions in the included articles as shown in Table IV. [18], [26], [16], [29], [19], [17], [21] Describe/analyze trade-off 5 [13], [22], [24], [2], [20] Solutions for single attributes 8 [33], [14], [23], [25], [15], [28], [30], [32] Solutions for trade-off 13 [34], [35], [36], [37], [38], [39], [27], [40], [41], [31], [42], [43], [44] The papers describing or analyzing single attributes concentrate on one or more of the quality attributes. We have seven such papers.…”

MPI to Coarray Fortran: Experiences with a CFD Solver for Unstructured Meshes

“…Haveraaen et al [23] observed that, even for large messages, the Intel compiler was performing element-wise transfer when Coarray Fortran was used for communication. similar (till 96 cores) or in some cases (192 and 384 cores) better than the MPI version.…”

A Monte Carlo-based Poisson's equation solver parallelized with Coarray Fortran

“…CAF's speedup performance was found to be 3.6% higher on average than the ideal value. This superlinear speedup is caused by cache effects [20]. More data can be stored in caches as the number of processors increases.…”