Accelerating MATLAB slow loop execution with CUDA

MATLAB´s high-level programming syntax and user-friendly environment makes it best suited to write technical code. MATLAB is extensively used in a number of scientific fields, such as digital image processing, digital signal processing, mathematics (just to name a few). The significance of MATLAB´s programming is its rich library of built-in functions that provide programming ease. However, MATLAB uses an interpreter which slows down the processing, especially while executing loops. This becomes a performance bottleneck in programs that make excessive use of loops. In the field of digital image processing, for example, the operations performed on a matrix usually involve many nested loop structures. Therefore, MATLAB is often exposed to the issues of slow execution and memory latency. In order to accelerate MATLAB´s processing, we utilize NVIDIA´s CUDA parallel processing architecture. We demonstrate that interfacing MATLAB with CUDA and parallelizing the most time-consuming portions of MATLAB´s code, the processing can be speeded up significantly. As a proof of concept, we present some results of JPEG compression with MATLAB using CUDA. Not only our results show a significant performance boost in MATLAB´s processing, but are also in perfect agreement with quality.