| 摘要 |
A memory access method for parallel computing, which is applied in the case that n (n=2) parallel threads invoke the same original execution module, comprises the following steps: S1) determining the number k of mirror execution modules needed to be created based on the number n of threads, wherein n-1=k=1; S2) determining the mapping policy F between the n threads and the execution modules including the original execution module and the k mirror execution modules; S3) creating k mirror execution modules, and allocating the individual private memory pool; S4) executing the n parallel threads, and making each thread invoke the execution module according to the mapping policy F, wherein each execution module only accesses its private memory pool; S5)deleting the created mirror execution modules after finishing the thread operation, and releasing the resources. The invention can enable the efficiency of parallel computing to linearly increase as the number of cores increases, without the limitation of the memory access bottleneck. The method is universal in various operating systems, and single-thread programs can be directly transplanted into a multi-thread environment conveniently. |
