Register Allocation for a Unified VLIW-based RISC/DSP Processor

碩士 === 國立中正大學 === 資訊工程所 === 95 === With the popularity of SOC, embedded processors with new technology emerged rapidly. To reduce the cost and improve performance simultaneously, all sorts of architecture have been investigated. Although architecture plays a significant role in the performance of pr...

Full description

Bibliographic Details
Main Authors: Min-Chin Hung, 洪敏進
Other Authors: Rong-Guey Chang
Format: Others
Language:en_US
Published: 2007
Online Access:http://ndltd.ncl.edu.tw/handle/78900684982326291459
id ndltd-TW-095CCU05392026
record_format oai_dc
spelling ndltd-TW-095CCU053920262015-10-13T14:08:36Z http://ndltd.ncl.edu.tw/handle/78900684982326291459 Register Allocation for a Unified VLIW-based RISC/DSP Processor 以VLIW為基礎的貼心處理器之暫存器配置 Min-Chin Hung 洪敏進 碩士 國立中正大學 資訊工程所 95 With the popularity of SOC, embedded processors with new technology emerged rapidly. To reduce the cost and improve performance simultaneously, all sorts of architecture have been investigated. Although architecture plays a significant role in the performance of processors, compiler is essential and crucial to it. Our proposed architecture has two cores and three register banks, and one of the register banks is shared by the two cores. The banked register file reduces the cost but results in more constraints and overhead, so that we desire to reduce the overhead by exploiting the shared register bank. Various register allocation algorithms have been proposed so far; however, none of them would perform well over all kinds of architectures. In this paper, we proposed a novel register allocation algorithm for our architecture. The algorithm takes instruction scheduling and code partitioning into consideration and adopts a heuristic analysis. To demonstrate the efficiency of our approach, we implemented it on GCC and simulated on a cycle accurate simulator. And the experimental results show that our proposed method has excellent performance improvement on average. Rong-Guey Chang 張榮貴 2007 學位論文 ; thesis 51 en_US
collection NDLTD
language en_US
format Others
sources NDLTD
description 碩士 === 國立中正大學 === 資訊工程所 === 95 === With the popularity of SOC, embedded processors with new technology emerged rapidly. To reduce the cost and improve performance simultaneously, all sorts of architecture have been investigated. Although architecture plays a significant role in the performance of processors, compiler is essential and crucial to it. Our proposed architecture has two cores and three register banks, and one of the register banks is shared by the two cores. The banked register file reduces the cost but results in more constraints and overhead, so that we desire to reduce the overhead by exploiting the shared register bank. Various register allocation algorithms have been proposed so far; however, none of them would perform well over all kinds of architectures. In this paper, we proposed a novel register allocation algorithm for our architecture. The algorithm takes instruction scheduling and code partitioning into consideration and adopts a heuristic analysis. To demonstrate the efficiency of our approach, we implemented it on GCC and simulated on a cycle accurate simulator. And the experimental results show that our proposed method has excellent performance improvement on average.
author2 Rong-Guey Chang
author_facet Rong-Guey Chang
Min-Chin Hung
洪敏進
author Min-Chin Hung
洪敏進
spellingShingle Min-Chin Hung
洪敏進
Register Allocation for a Unified VLIW-based RISC/DSP Processor
author_sort Min-Chin Hung
title Register Allocation for a Unified VLIW-based RISC/DSP Processor
title_short Register Allocation for a Unified VLIW-based RISC/DSP Processor
title_full Register Allocation for a Unified VLIW-based RISC/DSP Processor
title_fullStr Register Allocation for a Unified VLIW-based RISC/DSP Processor
title_full_unstemmed Register Allocation for a Unified VLIW-based RISC/DSP Processor
title_sort register allocation for a unified vliw-based risc/dsp processor
publishDate 2007
url http://ndltd.ncl.edu.tw/handle/78900684982326291459
work_keys_str_mv AT minchinhung registerallocationforaunifiedvliwbasedriscdspprocessor
AT hóngmǐnjìn registerallocationforaunifiedvliwbasedriscdspprocessor
AT minchinhung yǐvliwwèijīchǔdetiēxīnchùlǐqìzhīzàncúnqìpèizhì
AT hóngmǐnjìn yǐvliwwèijīchǔdetiēxīnchùlǐqìzhīzàncúnqìpèizhì
_version_ 1717748831534186496