Chris Mauzey

Since September of 2010, I have been finding ways to accelerate the execution of programs used to process the REAL data. Sometimes it's as simple as refactoring inefficient code. But, the largest and most exciting gains result from the deliberate use of vector processing capabilities of CPUs and massively parallel processing capabilities of graphical processing units (GPUs).

For example, by using the branchless vectorized median (BVM) filter algorithm, I was able to make our high-pass median filter ~70x faster than the IDL implementation. It was programmed in C and used Streaming SIMD Extensions and OpenMP; it ran on a 6-core 3.33GHz Xeon 5680. Further acceleration was gained when using the BVM algorithm implemented in CUDA C, which was almost twice as fast as the previous example when running on an nVidia Tesla C2070. I have also used the library GPULib from Tech-X Corp, which helps in refactoring IDL code to run on CUDA-enabled GPUs.

Since 2010, I have coauthored 4 journal articles and 13 conference presentations including the three below.

References

Mauzey, C. F., P. Dérian, and S. D. Mayor, 2014: Wavelet-based optical flow for real-time wind estimation using CUDA, Poster presentation 4253 in CV09. GPU Technology Conference (GTC), 24-27 March, San Jose, CA. Poster

Mauzey, C. F., J. P. Lowe, and S. D. Mayor, 2012: Real-time wind velocity estimation from aerosol lidar data using graphics hardware. Poster presentation AV10 at the GPU Technology Conference (GTC), 14-17 May 2012, San Jose, CA. Poster

Mauzey, C. F., J. P. Lowe, S. D. Mayor, 2011: The Use of Graphics Hardware in Computing Wind Velocity from Aerosol Data. College of Natural Sciences Poster Session. Poster

Back to Lidar Team.

Chris Mauzey working on computer