Understanding the Power of Evolutionary Computation for GPU Code Optimization

Achieving high performance for GPU codes requires developers to have significant knowledge in parallel programming and GPU architectures, and in-depth understanding of the application. This combination makes it challenging to find performance optimizations for GPU-based applications, especially in scientific computing. This paper shows that significant speedups can be achieved on two quite different scientific workloads using the tool, GEVO, to improve performance over human-optimized GPU code. GEVO uses evolutionary computation to find code edits that improve the runtime of a multiple sequence alignment kernel and a SARS-CoV-2 simulation by 28.9% and 29% respectively. Further, when GEVO begins with an early, unoptimized version of the sequence alignment program, it finds an impressive 30 times speedup-a performance improvement similar to that of the hand-tuned version. This work presents an in-depth analysis of the discovered optimizations, revealing that the primary sources of improvement vary across applications;that most of the optimizations generalize across GPU architectures;and that several of the most important optimizations involve significant code interdependencies. The results showcase the potential of automated program optimization tools to help reduce the optimization burden for scientific computing developers and enhance performance portability for domain-specific accelerators. © 2022 IEEE.

Boosting the analysis of protein interfaces with multiple interface string alignments: Illustration on the spikes of coronaviruses.

We introduce multiple interface string alignment (MISA), a visualization tool to display coherently various sequence and structure based statistics at protein-protein interfaces (SSE elements, buried surface area, ΔASA , B factor values, etc). The amino acids supporting these annotations are obtained from Voronoi interface models. The benefit of MISA is to collate annotated sequences of (homologous) chains found in different biological contexts, that is, bound with different partners or unbound. The aggregated views MISA/SSE, MISA/BSA, MISA/ΔASA, and so forth, make it trivial to identify commonalities and differences between chains, to infer key interface residues, and to understand where conformational changes occur upon binding. As such, they should prove of key relevance for knowledge-based annotations of protein databases such as the Protein Data Bank. Illustrations are provided on the receptor binding domain of coronaviruses, in complex with their cognate partner or (neutralizing) antibodies. MISA computed with a minimal number of structures complement and enrich findings previously reported. The corresponding package is available from the Structural Bioinformatics Library (http://sbl.inria.frand https://sbl.inria.fr/doc/Multiple_interface_string_alignment-user-manual.html).