OpenCL 2.0 Compiler Adaptation on LLVM for PTX Simulators

OpenCL continues to gather momentum on both desktop and mobile devices. The new features of OpenCL 2.0 provides developers better expressive power in programming heterogeneous computing environments. Currently in the experimental simulation environment, gem5-gpu only supports CUDA, but GPGPU-Sim can support OpenCL by compiling OpenCL kernel code to PTX using real GPU driver. However, this driver compilation in GPGPU-Sim only can support up to OpenCL 1.2. To support OpenCL 2.0, it is necessary to extend the compiler to enable the compilation of OpenCL 2.0 kernel code to PTX. In this paper, our experience in enabling the compiler flow is reported. In OpenCL 2.0, it provides new features such as dynamic parallelism, work-group built-in functions, extend atomic built-in functions, and so on. The proposed compiler that is modified from Low Level Virtual Machine (LLVM) extends such features for enhancing the emulator to support OpenCL 2.0. After the compiler is modified, it can support dynamic parallelism, work-group built-in functions and extend atomic built-in functions. Using existing dynamic parallelism APIs in CUDA to implement OpenCL 2.0 enqueue kernel and revise compilation scheme in clang. Furthermore, the proposed compiler also creates local buffers for each work group to use for work-group built-in functions, and adds atomic built-in functions with memory order and memory scope for OpenCL 2.0 in NVPTX. From benchmarks, the proposed compiler can support the claim target.