Dr. Jeremy Bennett brings us two short lectures: the first one is under the theme "Free softcores, tools and toolchains: The OpenRISC experience", the second short lecture is about "MAGEEC: Machine Guided Energy Efficient Compilation".
Free softcores, tools and toolchains: The OpenRISC experience:
In this talk we will look at the availability of free and open source softcores, EDA tools and compiler tool chains. Central to this will be a presentation of the OpenRISC 1000, a fully open 32/64-bit RISC processor architecture. Inspired by the MIPS and DLX architectures, the OpenRISC 1000 has many Verilog implementations and is used in a wide range of commercial products including Samsung set top boxes, NXP Jennic Zigbee chips and NASA's TechEdSat, which flew in 2012/13. In addition to the design and Verilog implementations being fully open, the processor is supported by open source front-end EDA tools such as Icarus Verilog and Verilator. It has a comprehensive and robust GNU tool chain, with an experimental LLVM tool chain also available. Linux and a wide-range of RTOS are supported. As well as describing the engineering implementation, the talk will look at how such an open design has been successful in a commercial environment, the business models that are most appropriate to such an open source approach, and where such business models can fail.
MAGEEC: Machine Guided Energy Efficient Compilation:
We are used to compilers which optimize for execution speed, and (in the embedded sector) for code size. In 2012 James Pallister of Bristol University and Embecosm led the seminal research project which demonstrated conclusively that compiler optimization has a major impact on the energy consumed by the generated code (http://comjnl.oxfordjournals.org/content/early/2013/11/11/comjnl.bxt129.abstract?keytype=ref&ijkey=aA4RYlYQLNVgkE3). This finding has immense potential for data center power usage, for battery life of consumer devices, for the efficiency of devices relying on energy scavenging and for remote sensing, where batteries must last for years at a time. In this short talk, we will explore how compiled programs consume energy and the opportunities for compiler optimization to reduce energy consumption. We will provide an introduction to MAGEEC, an 18-month project supported by the UK Technology Strategy Board, which uses machine learning to select compiler optimizations that will yield the most energy efficient compiled code.
Dr Jeremy Bennett is Embecosm’s founder, an expert on silicon chip modeling, source level debuggers and compilers, for which Embecosm provides commercial support services. A former academic, Jeremy holds a MA and PhD from Cambridge University and is a Member of the British Computer Society, Chartered Engineer, Chartered Information Technology Professional and Fellow of the Royal Society of Arts. He is the author of the standard textbook, “Introduction to Compiling Techniques” (McGraw-Hill 1990, 1996, 2003).
José João Henriques Teixeira de Sousa