Kalle Raiskila

pocl: A Performance-Portable OpenCL Implementation

OpenCL is a standard for parallel programming of heterogeneous systems. The benefits of a common programming standard are clear; multiple vendors can provide support for application descriptions written according to the standard, thus reducing the program porting effort. While the standard brings the obvious benefits of platform portability, the performance portability aspects are largely left to the programmer. The situation is made worse due to multiple proprietary vendor implementations with different characteristics, and, thus, required optimization strategies. In this paper, we propose an OpenCL implementation that is both portable and performance portable. At its core is a kernel compiler that can be used to exploit the data parallelism of OpenCL programs on multiple platforms with different parallel hardware styles. The kernel compiler is modularized to perform target-independent parallel region formation separately from the target-specific parallel mapping of the regions to enable support for various styles of fine-grained parallel resources such as subword SIMD extensions, SIMD datapaths and static multi-issue. Unlike previous similar techniques that work on the source level, the parallel region formation retains the information of the data parallelism using the LLVM IR and its metadata infrastructure. This data can be exploited by the later generic compiler passes for efficient parallelization. The proposed open source implementation of OpenCL is also platform portable, enabling OpenCL on a wide range of architectures, both already commercialized and on those that are still under research. The paper describes how the portability of the implementation is achieved. We test the two aspects to portability by utilizing the kernel compiler and the OpenCL implementation to run OpenCL applications in various platforms with different style of parallel resources. The results show that most of the benchmarked applications when compiled using pocl were faster or close to as fast as the best proprietary OpenCL implementation for the platform at hand.

Power optimizations for transport triggered SIMD processors

Power consumption in modern processor design is a key aspect. Optimizing the processor for power leads to direct savings in battery energy consumption in case of mobile devices. At the same time, many mobile applications demand high computational performance. In case of large scale computing, low power compute devices help in thermal design and in reducing the electricity bill. This paper presents a case study of a customized low power vector processor design that was synthesized on a 28 nm process technology. The processor has a programmer exposed datapath based on the transport triggered architecture programming model. The papers focus is on the RTL and microarchitecture level power optimizations applied to the design. Using semiautomated interconnection network and register file optimization algorithm, up to 27% of power savings were achieved. Using this as a baseline and applying register file datapath gating, register file banking and enabling clock gating of individual pipeline stages in pipelined function units, up to 26% of power and energy savings could be achieved with only a 3% area overhead. On top of this, for the measured radio applications, the exposed datapath architecture helped to achieve approximately 18% power improvement in comparison to a VLIW-like architecture by utilizing optimizations unique to transport triggered architectures.