Akira was born and raised on the island of Oahu. After graduating from Kaimuki High School, he went on to pursue a Bachelor of Science degree in Computer Engineering at the University of Hawai‘i at M?noa. Akira is also involved with the Hawaii Undergraduate Initiative Program (HUI), where he is able to work with incoming freshman to support and assist them in their transition from high school to college. Akira’s interest outside of academics are hiking, weight training, photography, and spending quality time with family and friends. He hopes to one day pursue a career in STEM in Hawai‘i.

Home Island: Oahu
High School: Kaimuki High School
Institution when accepted: University of Hawaii at Manoa

Akamai Project: Optimization for the Ray Tracing Algorithm for TMT Adaptive Optics Reconstruction Parameter Generator
Project Site: TIO/TMT International Observatory, LLC
Mentor: Lianqi Wang

Project Abstract:

The Adaptive Optics (AO) team from the TMT International Observatory is looking to maximize the performance and efficiency of the ray tracing algorithm that deals with large scale numerical computations. The Reconstruction Parameter Generator (RPG) relies on the ray tracing algorithm to calculate the parameters for AO wavefront reconstruction, in order to drive the deformable mirrors to correct the turbulence distorted wavefront. The ray tracing routine is a critical part of the wavefront reconstruction algorithm. Massive parallelization and efficient use of processing power is critical to meet timing requirements. The Intel Xeon Phi many-core CPU emerges as a viable architecture for TMT’s RPG. The computing efficiency of the ray tracing algorithm can be determined by calculating the floating-point operations per second (FLOPS) of the program and comparing it to the central processing units (CPUs) rated FLOPS. We compared the efficiency achieved using several parallel programming techniques from the Open Multi-Processing (OpenMP) Application Program Interface (API). Parallel programming has allowed the ray tracing function to execute many calculations simultaneously by dividing the large calculations into smaller calculations and letting them execute in parallel in many computing cores. To achieve peak efficiency, we examined the current algorithm and determined where most of the time is spent in the program. Then we improved the algorithm with the use of parallelization, which has reduced the computational time from 1.46 seconds down to 9.548 milliseconds. However, the ray tracing algorithms FLOPS is only 1% of the Intel Xeon Phi rated FLOPS, which means that the program does not fully maximize the processing power of the Intel Xeon Phi CPU. The ray tracing algorithms can be improved in both computational time and FLOPS by implementing scalar performance computing, vectorization with structure of arrays (SoA), and managing memory traffic efficiently. With these improvements, the RPG will be able to utilize the Intel’s 64 cores and 256 threads to compute the ray tracing algorithm in time and will allow the RPG to finish preparing reconstruction parameters in time for the remaining part of the AO system.