A Program Decompression Engine IP Generator Based on Partially Field-partitioned Compression Technique for the Embedded System

• Main Authors: Ko-Yen Hu, 胡格嚴
• Other Authors: 蔣元隆
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/61693611602232671821

碩士 === 崑山科技大學 === 數位生活科技研究所 === 98 === In this thesis, we bring up the partial field partitioning compression and the decompression system, and get better results to deal with the jumping instruction. According to the ARM’s jumping instructions, they are divided into direct jumping instruction and indirect jumping instruction. We propose the problem of the compressed cost according the repeated times in code, and depend on compression cost and jumping instruction to distinguish from all program codes; divided into 32-bits compressed instruction, 32-bits uncompressed instruction and differed instruction, and gather statistics of all the different blocks of code to compress all the different blocks by using Huffman Coding to reach a better total cost (decompression cost + storing cost). However, in this thesis, we propose a method that for these instructions which occur rare times to do some instruction compression. These instructions which only occur rare times of instructions occupy the number of total program instructions above 1/2. If leave these instructions uncompressed, it is cheaper than the compressed ones. (Because doing compression needs decompressed hardware.) But if cutting these instructions into two 16-bits fields, then perform field partitioning compression and add decompressed circuit will still match economic efficiency. But if we fully perform field partitioning compression and decompress all program instructions, the economic efficiency will be worse than only using 32-bits to compress and decompress. So my thesis proposes partial field partitioning compression method. First, we do 32-bits full compression, differed instruction compression, then, according to the uncompressed 32-bits, we do compression on front 16-bits and rear 16-bits; after that we compress and decompress them, it will contribute to reduce the cost.