I completed my Ph.D. in 2017 under the direction of Jeremy Buhler. My research interests broadly include parallel programming (esp. in SIMD) and acceleration of streaming applications on graphics cards (GPUs). Specifically, my main research focus has been on exploiting GPU parallelism for computationally “irregular” applications not traditionally well-suited to GPUs due to runtime-dependent data flow. I have also worked on accelerating short-read DNA sequence alignment on a GPU.
MERCATOR project
- Project page (coming soon)
- MERCATOR = Mapping EnumERATOR for CUDA
- Purpose: enable developers to easily write performant GPU code for modular streaming applications
- Method:
- automatically generate and provide data movement and kernel framework code, leaving only modular application code for developer to write
- pack input queues and schedule module execution so as to maintain high SIMD occupancy in the presence of runtime-dependent data flow
- Details: See dissertation and HPCS 2017 publication
Dissertation
Title: “Efficiently and Transparently Maintaining High SIMD Occupancy in the Presence of Wavefront Irregularity”
Publications
Cole, Stephen V. and Buhler, Jeremy. “MERCATOR: a GPGPU framework for Irregular Streaming Applications.” In IEEE 15th International Conference on High Performance Computing & Simulation (HPCS 2017), 2017 (to appear).
Cole, Stephen V. and Buhler, Jeremy D. “The Function-Centric Model: Supporting SIMD Execution of Streaming Computation.” In Proceedings of the Fifth Annual Workshop on Data Flow Models for Extreme-Scale Computing (DFM 2015), PACT ’15.
Cole, Stephen V. and Gardner, Jacob R. and Buhler, Jeremy D.. “WOODSTOCC: Extracting Latent Parallelism from a DNA Sequence Aligner on a GPU.” In IEEE 13th Int’l Symp. Parallel & Distributed Computing, 2014.