Performance evaluation of macrodataflow computation on shared memory multiprocessors
Kento Aida, Kousei Iwasaki, Hironori Kasahara, S. Narita · 2002
The coarse grain parallel processing on shared memory multiprocessor systems has been implemented using multi-tasking. However, this scheme has drawbacks such as difficulty in the extraction of parallelism among coarse grain tasks by ordinary users and large dynamic scheduling overhead caused by operating system calls or run-time library calls. On the other hand, in the proposed Fortran macrodataflow computation scheme, the compiler automatically generates coarse grain tasks called macrotasks, exploits parallelism among macrotasks, and generates a dynamic scheduling routine that schedules macrotasks to processors at run-time with small overhead. This paper presents performance evaluation of macrodataflow computation on shared memory multiprocessor systems. The results on four processors of a KSR1 show that macrodataflow computation reduces execution time to 1/2.81 of sequential execution time while an ordinary multi-threading reduces execution time to 1/2.19 of sequential execution time.