Helium: a transparent inter-kernel optimizer for OpenCL

Abstract

State of the art automatic optimization of OpenCL applications focuses on improving the performance of individual compute kernels. Programmers address opportunities for inter-kernel optimization in specific applications by ad-hoc hand tuning: manually fusing kernels together. However, the complexity of interactions between host and kernel code makes this approach weak or even unviable for applications involving more than a small number of kernel invocations or a highly dynamic control flow, leaving substantial potential opportunities unexplored. It also leads to an over complex, hard to maintain code base.We present Helium, a transparent OpenCL overlay which discovers, manipulates and exploits opportunities for inter-and intrakernel optimization. Helium is implemented as preloaded library and uses a delay-optimize-replay mechanism in which kernel calls are intercepted, collectively optimized, and then executed according to an improved execution plan. This allows us to benefit from composite optimizations, on large, dynamically complex applications, with no impact on the code base. Our results show that Helium obtains at least the same, and frequently even better performance, than carefully handtuned code. Helium outperforms handoptimized code where the exact dynamic composition of compute kernel cannot be known statically. In these cases, we demonstrate speedups of up to 3x over unoptimized code and an average speedup of 1.4x over hand optimized code.

Using Cisco Network Components to Improve NIDPS Performance

“…When some nodes experience a high volume of data, the buffer will fill up and packets may be dropped. Configuring the kernel parameter in the New Application Programming Interface (NAPI) can enhance kernel performance by increasing the level of optimization and selecting multivariate features such as kernel complex quantitative near-infrared (K-NIR), kernel support vector regression (k-SVR), or kernel partial least squares (K-PLS) to improve the accuracy of packet processing [23,24,25,26 ]. In order to hold and process packets quickly, these kernel performance enhancements pull a high value of packets from interfaces and bind them with obtainable CPU cycles, which limit packet speed and time and have no buffer memory.…”

Performance Optimisation of Parallelized ADAS Applications in FPGA-GPU Heterogeneous Systems: A Case Study With Lane Detection

“…In [29], the authors proposed a framework combining OpenCL application auto-tuning and runtime resource management. The study in [30] presented a transparent OpenCL overlay called Helium, for inter-and intra-kernel optimisation. The studies mentioned above are not yet mature and to the best of our knowledge, state-of-the-art research remains at the stage that optimisations are highly dependent on the specific algorithm, architecture, and programming features.…”

Improving the performance of ADAS application in heterogeneous context: A case of lane detection

“…In [7], the authors proposed a framework combing OpenCL application auto-tuning and runtime resource management. Authors in [8] presented a transparent OpenCL overlay called Helium, for inter-and intra-kernel optimization.…”