FLAP: Tool to generate CUDA code from sequential C code

Abstract

The GPU usage has increased in the high performance computing applications. However, still programming tools required to facilitate the development of programs for GPUs. It is well known that the loops in sequential programs are potential sources of parallelization. Three types of FOR-loops have been identified that can be mapped to CUDA kernel: single loops, nested loops and loops involving reduction operations. These loops can be mapped to CUDA kernels using patterns. In this work we present FLAP, a tool to generate CUDA parallel code from sequential C code. This tool uses these patterns to generate parallel code CUDA. We present some tests to show the performance of the code that generates FLAP.

Related Papers

• → Parallel connected-component labeling algorithm for GPGPU applications(2010)14 cited
• Parallel Programming For High-Performance Computing on CUDA(2009)
• CUDA-NP： Realizing Nested Thread-Level Parallelism in GPGPU Applications(2015)
• Introductory on GPGPU Programming Technique(2010)
• → Новітні архітектури відеоадаптерів. Технологія GPGPU. Частина 2(2013)