Application program and special instruction set integrated processor agile design method

Abstract

The invention provides an application program and special instruction set integrated processor agile design method. The method comprises the following steps: establishing an application and test environment of an application program to be operated on a processor; formulating a processor expansion architecture and an expansion description rule, representing basic expansion units in the processor byusing a specified data type, and representing an expansion instruction in the processor by using an expansion instruction definition function; compiling an expansion instruction definition function under the formulated processor expansion architecture and the expansion description rule; and realizing the hot spot part of the application program by using the expansion instruction definition function. The processor expansion instruction is described by adopting an advanced programming language, so that a general engineer can effectively master the processor expansion instruction. The definitionof the processor extension instruction only comprises an operation definition, and details for realizing the instruction in the processor micro-architecture are not involved. Integrated research anddevelopment of processor extension instructions and application programs are achieved, and a development environment in a rapid cycle of definition, use and verification of the expansion instructionsis achieved.

Description

technical field [0001] 本发明涉及半导体技术领域,具体地,涉及应用软件及专用指令集处理器一体化敏捷设计方法。 Background technique [0002] 集成电路设计方法需要不断更新才能跟得上摩尔定律。目前主流芯片设计所采用的方法基于硬件描述语言(HDL),如Verilog或VHDL,发明已有三十多年。在这期间,半导体生产技术和集成电路的复杂度增加了近乎100万倍。传统HDL设计手段已经明显跟不上半导体制造工艺发展速度,不能满足市场需求。如何才能改善设计效率,更快更好更灵活地设计集成电路是一个半导体行业一直在探讨的问题,有着巨大的市场需求和商业潜力。 [0003] 集成电路设计方法有两个非常重要但相互矛盾的指标:运算效率和设计灵活性。超高的运算效率是半导体芯片在商业价值所在。而高速发展变换的市场环境又需要芯片设计具有很高的灵活性。实现高效运算需要在芯片电路里添加很多和应用有关的运算单元(算子),并且设计相应的控制电路来确保数据正确地使用这些算子。这种类型的设计方法统称硬件设计,几乎都是用HDL来描述所需电路,依赖昂贵的计算机辅助设计(CAD)工具来合成电路并对电路继续仿真以确保其正确性。传统硬件设计方法虽然能满足电路高效的运算需求,但设计周期很长,代价高,验证困难,修改麻烦。在集成电路里,还有一种完成计算任务的方法,那就是在芯片里植入一个或多个通用处理器。通过软件编程的方式在处理器上完成特定的运算。这种软件设计方式非常灵活,修改起来容易,查找问题简单,甚至在已经生产好的芯片上也能通过软件来更改算法。但软件设计最大的缺点是运算效率非常低,被通用处理器的指令集所限制。通用处理器的指令集都是一些非常简单的运算。复杂的运算需要通过很多条指令才能完成。总结起来,硬件设计方法能实现高效运算,但设计效率低。软件设计正好相反,设计效率高,但运算效率低。 [0004] 为了解决硬件和软件设计这一对矛盾,取两者之长而避两者之短,越来越多的设计采用一种混合型设计方式:在通用处理器上针对应用的特殊算法添加具有很强运算能力的专用指令。这种方式既保留的软件设计的灵活性,又能同时达到很高的运算效率。其结果是专门为某一个或一类应用而设计的处理器,又称“应用专属指令集处理器(ASIP)”,或简称“定制处理器”。定制处理器的设计工具和...

AI Technical Summary

Problems solved by technology

Inability to implement all custom processor behaviors with an efficient microframework, resulting in suboptimal performance of the resulting custom processor

[0007] 2) The custom processor definition and algorithm optimization are disconnected and completed in two different environments, resulting in a long design cycle and very inefficient troubleshooting methods

[0008] 3) Engineers who write custom processor description code must have deep knowledge of processor architecture, so they cannot be accepted by most software and algorithm engineers

[0013] 2) The description language used for customization is out of touch with the language used by application software, which brings great trouble to verification, because the cause of the problem may be in different places

[0014] 3) From the description of extended instructions, to the generation of software tools, and then to software optimization, the cycle of the whole link is too long, which affects the design efficiency

Images