lect4_华科并行编程课件.pdf

Parallel Programming Principle and Practice Lecture 4 — Parallel Programming Methodology Jin, Hai School of Computer Science and Technology Huazhong University of Science and Technology Outline  Motivating Problems  Steps in Creating a Parallel Program  What a Simple Parallel Program Looks Like 2 Parallel programming methodology MOTIVATING PROBLEMS 3 Motivating Problems  Simulating Ocean Currents  Regular structure, scientific computing  Simulating the Evolution of Galaxies  Irregular structure, scientific computing  Rendering Scenes by Ray Tracing  Irregular structure, computer graphics 4 Simulating Ocean Currents  Model as two-dimensional grids  Discretize in space and time  finer spatial and temporal resolution - greater accuracy  Many different computations per time step  set up and solve equations  Concurrency across and within grid computations 5 Simulating Galaxy Evolution  Simulate the interactions of many stars evolving over time  Computing forces is expensive 2  O(n ) brute force approach  Hierarchical methods take advantage of force law: G  Many time-steps, plenty of concurrency across stars within one 6 Rendering Scenes by Ray Tracing  Shoot rays into scene through pixels in image plane  Follow their paths  They bounce around as they strike objects  They generate new rays: ray tree per input ray  Result is color and opacity for that pixel  Parallelis