PhD Seminar • Computer Graphics | Scientific Computing — The Need for Speed: Making Existing Industrial Fluid Simulation Pipelines Faster with Strategic Adaptivity

Friday, May 22, 2020 12:30 pm - 12:30 pm EDT (GMT -04:00)

Please note: This PhD seminar will be given online.

Ryan Goldade, PhD candidate
David R. Cheriton School of Computer Science

In an effort to make fluid simulations faster, we propose to solve for pressure forces (a common bottleneck) using adaptive resolution grids. Although foundational methods for employing adaptivity are over a decade old, hardware limitations of the time and inefficient data structure design ultimately led to these methods being overlooked in industry. Recently, improved data structure design and substantial hardware improvements have reignited interest in the potential of adaptive methods. However, incorporating adaptivity into an existing uniform grid fluid simulation pipeline inherently requires the entire pipeline to be redesigned from the ground up. 

Instead, we propose to strategically apply adaptivity to solve for pressure forces as a drop-in replacement for an existing uniform grid pipeline. We carefully transfer data between the uniform and adaptive settings that offers the performance improvements expected from adaptivity without the daunting requirement of rewriting the entire pipeline. Our proposed adaptive method achieves a 7X performance improvement when solving for pressure forces compared with the uniform grid method and offers a 3X performance improvement for the entire simulation pipeline.

To join this PhD seminar online, please go to https://uwaterloo.webex.com/uwaterloo/j.php?MTID=m2234ef5241ded1fb3ca0c8bae24e51d3.