https://scholars.lib.ntu.edu.tw/handle/123456789/484478
標題: | ARES-architecture reinforcing superscalar | 作者: | Lin, Y.-H. Chang, M.-C. FEI-PEI LAI |
公開日期: | 1991 | 起(迄)頁: | 338-343 | 來源出版物: | International Symposium on VLSI Technology, Systems, and Applications, Proceedings | 摘要: | In the ARES, there are four major features. First of all, the ARES utilizes both static and dynamic scheduling methods. The instruction identifier bits (IDBs) are attached to each instruction, except jump and branch, to indicate which basic block (BB) the instruction originally belongs to. Thus, the compiler can move instructions across the boundaries of BBs to get more instruction level parallelism. Secondly, the separate architectures memory address calculation from memory access for store instruction. This divides the original two memory cycles of Store instruction for write-back cache into two separate cycles. Thirdly, the system divides the instruction of compare-and-branch into two steps: compare and test-then-branch. With this scheme and branch registers, one can combine the current BB with the following BBs (taken/untaken or both) into one BB to increase the schedulable instructions. Finally, the system follows the same way as the IFU of the MARS system to peep and absorb jump, so there is no delayed slot for jump in the ARES. The architecture can have a 1.62 speedup, compared with the MIPS-X, with a simple extra hardware support. © 1991 IEEE. |
URI: | https://scholars.lib.ntu.edu.tw/handle/123456789/484478 | ISSN: | 19308868 | DOI: | 10.1109/VTSA.1991.246736 | SDG/關鍵字: | Cache memory; VLSI circuits; Basic blocks; Hardware supports; Instruction level parallelism; Memory access; Memory address; Memory cycles; Static and dynamic scheduling; Superscalar; Memory architecture |
顯示於: | 生醫電子與資訊學研究所 |
在 IR 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。