Architecture Design Trade-offs among VLIW SIMD and Multi-core Schemes
Yaohua Wang, Shuming Chen, Kai Zhang, Hu Chen, Xiaowen Chen
Abstract
Yaohua Wang, Shuming Chen, Kai Zhang, Hu Chen, Xiaowen Chen
Abstract
Hybrid architectures combined of VLIW, SIMD and multi-core schemes are increasingly prevailing in media processors, due to the abundant parallelism existed in media applications. However, parameters for current combinations such as the VLIW length, SIMD width and core count are set mainly according to simple profiling or the designer's experience rather than a systematic and in-depth investigation. With the development of both the application and the design technologies, a large amount of parallelism, together with better availability of computation resources motivate the need for deeper insight into the trade-offs among VLIW, SIMD and multi-core schemes. In this paper, an analytical model is proposed for these combinational architectures, furthermore, the area cost and cycle time variation are also provided in the form of different design constraints, under which, decent trade-offs are obtained, providing valuable insights into the design of media processors.
OpenAlex reports 2 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Hybrid architectures combined of VLIW, SIMD and multi-core schemes are increasingly prevailing in media processors, due to the abundant parallelism existed in media applications. However, parameters for current combinations such as the VLIW length, SIMD width and core count are set mainly according to simple profiling or the designer's experience rather than a systematic and in-depth investigation. With the development of both the application and the design technologies, a large amount of parallelism, together with better availability of computation resources motivate the need for deeper insight into the trade-offs among VLIW, SIMD and multi-core schemes. In this paper, an analytical model is proposed for these combinational architectures, furthermore, the area cost and cycle time variation are also provided in the form of different design constraints, under which, decent trade-offs are obtained, providing valuable insights into the design of media processors.
Key concepts: SIMD, Very long instruction word, Computer science, Parallel computing, Computer architecture, Parallelism (grammar), Profiling (computer programming), Computation