Developed scripts and processes to improve LLVM compiler development and turnaround timebetween making changes to optimizations and performance characterizationDeveloped tools to combine profiles, assembly, and basic block counts to produce a single annotatedassembly diff between two versions of a binaryCharacterized performance of benchmarks using statistical analysis of results and profiles, deliveredreports to compiler developers and managers for use in internal reportsProduced comparative analysis reports for GCC and LLVMWorked with common industry benchmarks like SPEC2000, 2006, and 2017Produced sysroots and release compilers for team and internal releases

Co-designed and implemented GUI to interface with Matlab back-end created by facultySimulated mechanical engineering problems with Matlab: trusses, frames, mechanismsThis work was considered exemplary, so that our team was invited to present it at a Matlab conference (Curriculum Conference, June 2014).

Designed and implemented a database to store scores from tests given to undergraduate students to be used in the NCSU MAE department's next ABET report.Developed a program to read in raw data to the database, run queries, and output the aggregated results for use in the report.This work was also used in generating statistics for internal reports

Characterized the effect of compiler options on quality of generated code for an internal compiler-development team (LLVM-based). This compiler generates ARM and Thumb code (ARMv7 ISA). I documented the performance delta attained by each optimization flag.Analyzed benchmark performance, and identified segments that could be sped up with compiler optmizations. Documented each benchmark, and their most important (hot) traces and functions.This work was performed in a Linux environment. Benchmark code was C and C++, and included SPEC2000, SPEC2006, and Linpack. I created shell scripts to manage the data generated from this exercise.

Assisted professional technical support staff in repairing computers afflicted with malware, and assisted teachers and students in operating classroom technologies such as classroom computers, projectors, and A/V equipment.

Located areas of well-known benchmarks, such as SunSpider, that performed less than optimally on a processor being developed (Qualcomm's Krait CPU). Utilized CPU performance monitors and debug tools (gdb) to isolate these areas.Analyzed code to understand the essential algorithm being performed. This work involved understanding disassembled ARM code, as well as referring back to source code, when available.Created targeted benchmarks for CPU design team to analyze slowdowns. These benchmarks captured the essence of the issue, but were made to execute in much less time than the original workload. I utilized a combination of ARM assembly code, JavaScript, and C/C++ to create these benchmarks. I developed approximately fifteen of them.