基於位元遮罩的編碼壓縮兼考量功率消耗

• Main Author: 江英傑
• Other Authors: 陳伯岳
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/21880594480937239272

碩士 === 國立彰化師範大學 === 資訊工程學系 === 98 === Abstract Code compression is often discussed in the research of embedded systems recently. It exploits that the instructions’ length of embedded systems is fixed. By replacing instructions with shorter length, the program size will be smaller and the power consumption will be smaller, too. Among various compression schemes, dictionary-based ones are applied for their good compression ratios and fast decompression engines. To replace instructions contained in a selected dictionary by index will make the program size smaller. Bitmask-based code compression derived from the dictionary-based ones has a superior compression ratio and a fast decompression engine as well. It uses bitmasks to record difference between instructions in order to achieve lower compression ratio. In this thesis, we adopt the bitmask-based scheme and then take power consumption into account for a cacheless architecture. Replace some of its dictionary entries to achieve large reduction of power consumption while keeping a good compression ratio. Since bitmask-based code compression is of variable length, the power consumption comes from access of executed instruction and waste of instructions branching. We propose three approaches, namely by-access-saving, by-frequency and by-block, to replace part of the dictionary. The by-access-saving approach is to replace the dictionary entry one-by-one. The greatest access-saving instruction which is not in dictionary is to replace the minimum bit-saving instruction in dictionary. The by-frequency approach is similar except that the replacing instruction is the one with the highest frequency. The by-block approach replaces many instructions with similar executing frequencies at a time. For the overhead resulted from instructions branching, we develop an auxiliary post-procedure called by-alignment to eliminate the waste of access. Without altering the content of dictionary, it aligns some high frequency executed instructions’ positions to the very beginning of a memory access. According to the result of experiments, by-block with by-alignment provides the best result. In the best case, only 1.61% of compression ratio loss can result in 43.75% of power consumption ratio gain.