Maximize parallelism and minimize communication overheads are important issues for distributed memory systems. Communication and data redistribution cannot be avoided even when considering global optimization of data ...Maximize parallelism and minimize communication overheads are important issues for distributed memory systems. Communication and data redistribution cannot be avoided even when considering global optimization of data distribution and computation decomposition. A new approach based on loop fusion is presented exploiting pipelining parallelism, thus communication overhead can be hidden and data redistribution can be avoided. This technique exploits pipelining from complex loop structures, which distinguishes itself from traditional pipelining techniques. Ex-periments show that the technique is superior to other optimizations.展开更多
Empirical observations suggest that many instructions and groups of instructions having the same inputs,and producing the same outputs. Such instructions do not have to be executed repeatedly-their result can be obtai...Empirical observations suggest that many instructions and groups of instructions having the same inputs,and producing the same outputs. Such instructions do not have to be executed repeatedly-their result can be obtained from records where they have been saved previously. This paper proposes an approach that uses compiler technique to exploit reuse for instruction group. In this approach ,the compiler first identifies code regions whose computation can be reused during dynamic execution. For each region,a library routing is invoked before the instruction in the region is executed. The function of the library routes is to record several instances of the region execution,including input register values and output register values. When it is found that a region has an instance that can be reused,then all the instructions in the region can be skipped and hence improve the performance.展开更多
文摘Maximize parallelism and minimize communication overheads are important issues for distributed memory systems. Communication and data redistribution cannot be avoided even when considering global optimization of data distribution and computation decomposition. A new approach based on loop fusion is presented exploiting pipelining parallelism, thus communication overhead can be hidden and data redistribution can be avoided. This technique exploits pipelining from complex loop structures, which distinguishes itself from traditional pipelining techniques. Ex-periments show that the technique is superior to other optimizations.
文摘Empirical observations suggest that many instructions and groups of instructions having the same inputs,and producing the same outputs. Such instructions do not have to be executed repeatedly-their result can be obtained from records where they have been saved previously. This paper proposes an approach that uses compiler technique to exploit reuse for instruction group. In this approach ,the compiler first identifies code regions whose computation can be reused during dynamic execution. For each region,a library routing is invoked before the instruction in the region is executed. The function of the library routes is to record several instances of the region execution,including input register values and output register values. When it is found that a region has an instance that can be reused,then all the instructions in the region can be skipped and hence improve the performance.
