Divergence Reduction In GPUs With SIMT Architecture

Branch Divergence has a significant impact on the performance of GPU programs. I used three Novel Software based divergence reduction techniques namely Fixed Scheduling, Frequency scheduling, and Balanced scheduling that aims to reduce branch divergence and comparing the execution time of each schedules. The calculation of End Semester Examination marks of Engineering and MBA students is the application to which the scheduling techniques are applied. The divergence condition checked whether the student belongs to Engineering Section or MBA section and based on that calculations further processing have been carried out. Evaluation shows frequency scheduling and balanced scheduling exhibiting large improvement in execution time.