异构计算以ARM架构为例.pdf

Heterogeneous Computing in ARM Architecture Alan Tsai Business Development Manager Sept 2012 Agenda  Trends in Heterogeneous Computing  GPU Computing with ARM® Mali™-T600 series as example  Heterogeneous System Architecture™ (HSA) is the future 2 Trends in the Industry  Heterogeneous multiprocessing  Established approach for SoC design  Mix of many specialized accelerators, implementing different ISAs  Diverse programming approaches lead to lack of portability  Parallel computation for performance and efficiency  Endorsed at all levels of computer architecture  Parallel programming traditionally difficult  General purpose programmability of GPUs  Massive parallel computation potential  Increasing programmability 3 What is Parallel Computing?  Simply, doing multiple tasks simultaneously  Task-Parallel computing does different tasks concurrently  Reading email, playing music, and surfing the web are all separate tasks  In a multicore system, these can execute simultaneously  Data-Parallel computing does the same operation on a collection of data concurrently  Adjusting the contrast of the pixels of an image  Each thread executes the same code but with different data  Classic SIMD (single-instruction, multiple-data)  GPU computing is perfect for data-parallel applications 4 What is Heterogeneous Computing? GPU used as computational accelerators or companion processors CPU GPU Massively parallel architecture gives great