《多处理器并行软件开发经验之谈》由会员分享，可在线阅读，更多相关《多处理器并行软件开发经验之谈（25页珍藏版）》请在金锄头文库上搜索。
1、3 Software Development for Parallel and Multi Core Processing Kenn R Luecke The Boeing Company USA 1 Introduction The embedded software industry wants microprocessors with increased computing functionality that maintains or reduces space weight and power SWaP Single core processors were the key embe
2、dded industry solution between 1980 and 2000 when large performance increases were being achieved on a yearly basis and were fulfilling the prophecy of Moore s Law Moore s Law states that the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every
3、two years 1 With the increased transistors came microprocessors with greater computing throughput while space weight and power were decreasing However this free lunch did not last forever 2 The additional power required for greater performance improvements became too great starting in 2000 Hence sin
4、gle core microprocessors are no longer an optimal solution Although distributed and parallel programming solutions provide greater throughput these solutions unfortunately increase SWaP The most likely solution is multi core processors which have been introduced into the embedded processor markets M
5、ost microprocessor manufacturers have converted from developing single core processors to multi core processors With this conversion the prophecy of Moore s Law is still being achieved See Figure 1 and notice how the single core processors are not keeping pace with the multi core processors Multi co
6、re processors increase throughput while maintaining or reducing SWaP for embedded environments which make them a good hardware solution for the aerospace industry Intel in particular has estimated that by 2011 95 of all the processors it ships will contain a multi core design However the software ma
7、rket shows less optimism with multi core processors For instance only 40 of software vendors thought their tools would be ready for multi core processing by 2011 The reasons for software engineering s lack of excitement with multi core processors include the following drawbacks Lack and immaturity o
8、f multi core specific development and debug software tools Lack of multi core processor standards Lack and immaturity of multi core enabled system software 1 http en wikipedia org wiki Moore s law 2 Sutter H March 2005 The free lunch is over A fundamental turn toward concurrency in software Dr Dobb
9、s Journal Volume 30 Number 3 Embedded Systems High Performance Systems Applications and Projects 36 Lack of parallel programming experience by the software community Lack of parallel programming models to support these multi core processors An abundance of differentiated multi core processors from m
10、ultiple suppliers Greater differentiation with inexperience can be problematic for software developers converting applications for multi core processors Fig 1 Processor Transistor Counts and Moore s Law3 These problems led Chuck Moore a Senior Fellow at AMD to state To make effective use of Multi co
11、re hardware today you need a PhD in computer science 4 Therefore multi core software development has fallen behind multi core hardware development This chapter will provide information on the current best technologies tools methodologies programming languages models and frameworks for software devel
12、opment on multi core processors Where different software development options exist comparisons and recommendations will be provided to the reader 2 Multicore definition Previous multiprocessing as opposed to multi core processing solutions such as parallel and distributed programming involved two or
13、 more processors which doubled tripled or 3 Fittes Dale October 30 2009 Using Multicore Processors in Embedded Systems Part 1 EE Times 4 Moore Chuck May 12 2008 Solving the Multi core Programming Problem Dr Dobbs Journal Software Development for Parallel and Multi Core Processing 37 even quadrupled
14、the board space weight and power SWaP consumed and heat generated by the processing system These solutions could comprise large networks leading to data latencies between processing components However multi core processors place multiple processing cores on a single chip to increase processing power
15、 without noticeably increasing the system s SWaP and heat dissipation Also with multiple cores on a single chip the data latencies of distributed programming are mostly negated With multi core processing the computer industry continues pushing the performance power envelope through parallel processi
16、ng rather than increasing the processor clock speed For the most part serial computing has been the standard software development model with multiple cores on a processor now parallel computing is emerging as the new standard and very few programmers are well versed in parallel computing A multi core processor in general appears similar to the dual core and quad core processors displayed in Figure 2 In both cases each core has an associated L1 cache while the L2 cache is shared between all the c
