An iteration partition approach for cache or local memory thrashing on parallel processing

Abstract

Parallel processing systems with cache or local memory in the memory hierarchies have become very common. These systems have large-size cache or local memory in each processor and usually employ copy-back protocol for the cache coherence. In such systems, a problem called “cache or local memory thrashing” may arise in executions of parallel programs, when the data unnecessarily moves back and forth between the caches or local memories in different processors. The techniques associated with parallel compilers to solve the problem are not completely developed.

In this paper we present an approach to eliminate unnecessary data moving between the caches or local memories for nested parallel loops. This approach is based on relations between array element accesses and enclosed loop indexes in the nested parallel loops. The relations can be used to assign processors to execute the appropriate iterations for parallel loops in the loop nests with respect to the data in their caches or local memories. An algorithm to calculate the correct iteration of the parallel loop in terms of loop indexes of the previous iterations executed in the processor is presented in the paper, even though there is more than one subscript expression of the same array variable in the loop.

This method benefits parallel code with nested loop constructs in a wide range of applications, in which the array elements are repeatedly referenced in the parallel loops. The experimental results show that the technique is extremely effective—capable of achieving double speedups over application programs such as unpack benchmarks.

Supported by the National Science Foundation under grant no. MIP 8809328.