Towards Automatic Injection of Optimal Communication into Ocean Models
Tuesday, June 10, 2025 3:00 PM to Thursday, June 12, 2025 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Research Poster
Compiler and Tools for Parallel ProgrammingEarth, Climate and Weather ModelingPerformance Tools and Simulators
Information
Poster is on display and will be presented at the poster pitch session.
Ocean models have very high computational demands, but typically do not reach peak performance on modern HPC systems. Parallel execution on large distributed HPC systems also requires distributed communication, which significantly affects their overall performance but can be difficult to efficiently implement.
With “Poseidon”, an HPC-oriented source-to-source translator for Fortran-based fluid dynamics solvers used in ocean and weather models with regular grid structure,s we can take the sequential code automatically translate it into parallel Fortran code for modern distributed HPC systems.
We demonstrate Poseidon's approach using a research code implementing the 2D fast barotropic solver of full 3D ocean simulation models, which involves over 20 stencil-like kernels. After extracting information from the code into a Data Flow Graph (DFG), we can apply graph-based operations to modify computation and communications. The communication operations are local operations that work on single communication nodes in the DFG but take global dependencies into account. Our prototype for optimizing communication can infer the minimal required loop bounds, halo sizes and halo exchanges and inject corresponding communication nodes into the graph. Poseidon then uses the graph to generate distributed parallel code that can successfully be compiled and run on a distributed HPC system.
In future work, Poseidon automatic code rewrite should help to: port existing code to GPU, hide process communications latency and apply automatic differentiation.
Contributors:
Ocean models have very high computational demands, but typically do not reach peak performance on modern HPC systems. Parallel execution on large distributed HPC systems also requires distributed communication, which significantly affects their overall performance but can be difficult to efficiently implement.
With “Poseidon”, an HPC-oriented source-to-source translator for Fortran-based fluid dynamics solvers used in ocean and weather models with regular grid structure,s we can take the sequential code automatically translate it into parallel Fortran code for modern distributed HPC systems.
We demonstrate Poseidon's approach using a research code implementing the 2D fast barotropic solver of full 3D ocean simulation models, which involves over 20 stencil-like kernels. After extracting information from the code into a Data Flow Graph (DFG), we can apply graph-based operations to modify computation and communications. The communication operations are local operations that work on single communication nodes in the DFG but take global dependencies into account. Our prototype for optimizing communication can infer the minimal required loop bounds, halo sizes and halo exchanges and inject corresponding communication nodes into the graph. Poseidon then uses the graph to generate distributed parallel code that can successfully be compiled and run on a distributed HPC system.
In future work, Poseidon automatic code rewrite should help to: port existing code to GPU, hide process communications latency and apply automatic differentiation.
Contributors:
Format
On DemandOn Site
