high performance computing on graphics processing units: hgpu.org

GPUs critically rely on a complex system software stack comprising kernel- and user-space drivers and JIT compilers. Yet, existing GPU simulators typically abstract away details of the software stack and GPU instruction set. Partly, this is because GPU vendors rarely release sufficient information about their latest GPU products. However, this is also due to the lack of an integrated CPU-GPU simulation framework, which is complete and powerful enough to drive the complex GPU software environment. This has led to a situation where research on GPU architectures and compilers is largely based on outdated or greatly simplified architectures and software stacks, undermining the validity of the generated results. Making the situation even more dire, existing GPU simulation efforts are concentrated around desktop GPUs, making infrastructure for modelling mobile GPUs virtually non-existent, despite their surging importance in the GPU market. Still, mobile GPU designers are faced with the challenge of evaluating design alternatives involving hundreds of architectural configuration options and micro-architectural improvements under tight time-to-market constraints, to which currently employed design flows involving detailed, but slow simulations are not well suited. In this thesis we develop a full-system simulation environment for a mobile platform, which enables users to run a complete and unmodified software stack for a state-of-the-art mobile Arm CPU and Mali Bifrost GPU powered device, achieving 100% architectural accuracy across all available toolchains. We demonstrate the capability of our GPU simulation framework through a number of case studies exploring modern, mobile GPU applications, and optimize them using functional simulation statistics, unavailable with other approaches or hardware. Furthermore, we develop a trace-based performance model, allowing architects to rapidly model GPU configurations in early design space exploration.