To speed up a library call on a parallel computer.
When a user program is executed by one master processor 1 and only the internal part of a library routine is executed by plural slave processors 3 on the parallel computer, the slave processors 3 are started only once at the beginning and they are made to wait in infinite loops 9. At the time of the library call 2, a command and data required for execution are transmitted from the master processor by communication and data are distributed 5. Then, the slave processors are set to the same state as that before execution, namely, a communication waiting state from the master processor in the infinite loops 9 after parallel execution. Thus, the overhead of the starting of the slave processors can be reduced when the library is called plural times and the library call on the parallel computer can be speeded up.
ADACHI HITOSHI
HIRAO MASAHIKO
HITACHI TOHOKU SOFTWARE
Next Patent: FAIL-FIRST, FAIL-FUNCTIONAL, AND FAULT-TOLERANT-MULTIPROCESSOR-SYSTEM