Branch Merging for Effective Chen Exploitation Yunn-Yen of Instruction-Level Chen Chung-Ta King Parallelism* Chien-Ming Department of Computer Science National Tsing Hua University Hsinchu, Taiwan 30043, R.O.C. Abstract In this paper, we propose a novel algorithm for merging branches in VLIW and superscalar architectures employing multi-way branch mechanisms. The branch merging problem is to find sets of branches that, can be merged and executed concurrently. The advantages of using branch merging are: (1) to reduce the number of stalls due to branches and (2) to enlarge the size of basic blocks, We show that the general branch merging problem is NP-complete, and then propose a heuristic algorithm to solve this problem. high frequency of branches [6]. Thus, the instructionlevel parallelism exploitable within a basic block is also small, often limited to a factor of two to three. Such a degree of parallelism is not enough to fully utilize the computing resources available in a VLIW or superscalar processor. It is necessary to get more parallelism for processor performance [2, 4, 51. Architecture supporting multi-way branches is an effective approach [3] to decreasing the number of branches in a program. The penalty due to branches is thus reduced. Also, it
/lp/association-for-computing-machinery/branch-merging-for-effective-exploitation-of-instruction-level-QpoUUdZ34I