Bayesian Phylogenetic Analysis on Multi-Core Compute Architectures: Implementation and Evaluation of BEAGLE in RevBayes With MPI.

The article focuses on improving the computational efficiency of likelihood calculations in Bayesian phylogenetic inference by integrating the high-performance phylogenetic library BEAGLE into the software RevBayes. This integration, termed RevBayes+BEAGLE, enables multi-threading on multi-core CPUs and GPUs, utilizing hardware-specific vectorized instructions, while retaining RevBayes’ flexibility. Additionally, a native parallelization approach using the message passing interface (MPI), called RevBayes+MPI, was implemented. Benchmarking showed that RevBayes+BEAGLE offers significant speedups on single cores and GPUs, whereas RevBayes+MPI scales better with multiple CPU cores, with speedups depending mainly on alignment size and data type rather than tree size. The study also evaluated strategies for rescaling partial likelihoods to avoid numerical underflow and introduced a branch-based partial likelihood storing method that can accelerate computations at the cost of increased memory usage. These developments provide users with flexible options for efficient Bayesian phylogenetic analyses across diverse hardware configurations.