Shodor

This module teaches: 1) Conway's Game of Life as an example of a cellular automaton, 2) how cellular automata are used in solutions to scientific problems, 3) how to implement parallel code for Conway's Game of Life (including versions that use shared memory via OpenMP, distributed memory via the Message Passing Interface (MPI), and hybrid via a combination of OpenMP and MPI), 4) how to measure the performance and scaling of a parallel application in multicore and manycore environments, and 5) how cellular automata fall into the Structured Grid "dwarf" (a class of algorithms that have similar communication and computation patterns). Upon completion of this module, students should be able to: 1) Understand the importance of cellular automata in modeling scientific problems, 2) Design a parallel algorithm and implement it using OpenMP and/or MPI, 3) Measure the scalability of a parallel code over multiple or many cores, and 4) Explain the communication and computation patterns of the Structured Grid dwarf. It is assumed that students will have prerequisite experience with C or Fortran 90, *nix systems, and modular arithmetic.

This module introduces the basics of cellular automaton simulation with an application to studying the effect of fencing artificial watering points on adult cane toad invasion in Australia.

This module is largely stand-alone. It is "Part II" only in the sense that it does not contain the overview of dynamic programming seen in Part I, and does not recapitulate the introduction to CUDA. We will continue to refer the reader to various NVIDIA references where appropriate, particularly the NVIDIA CUDA C Programming Guide, and the CUDA API Reference Manual, and where we introduce new CUDA-specific ideas, will linger a bit longer by way of introduction. The algorithms described here are completely independent of Part I, so that a reader who already has some familiarity with CUDA and dynamic programming may begin with this module with little difficulty.

This module teaches matrix multiplication in the context of enumerating paths in a graph and the basics of programming in CUDA. It emphasizes the power of using shared memory when programming on GPGPU architectures.

This module explores the inner workings of the BLAST similarity search tool, considering the algorithm and the impact of various search conditions and settings on performance. Various approaches to parallelizing the computation and their performance impacts are considered. Benchmarking of the mpiBLAST parallel code is carried out at different scales.