Near Fine Grain Parallel Processing using a Multiprocessor with MAPLE

T. Abe, K. Iwai, T. Morimura, R. Ogawa, K. Yasufuku, and H. Amano (Japan)


: processor, cache, static scheduling, multi grain parallelism, parallel computing system


Multi-grain parallelizing scheme is one of effective par allelizing schemes which exploits various level paral lelism: coarse-grain(macro-dataflow), medium-grain(loop level parallelizing) and near-fine-grain(statements paral lelizing) from a sequential program. A multi-processor ASCA is designed for efficient execution of multi-grain parallelizing program. A processing element called MAPLE are mainly de signed for near-fine-grain parallelism, and has two mod ules called MAPLE core and DTC. The MAPLE core is a simple RISC processor which executes every operation in a fixed time and realize direct register to register transfer. The DTC realize a software controlled cache by instruc tions which are generated by the compiler. With a static scheduling, near-fine-grain parallel processing is efficiently performed using a communication mechanism with receive registers, and non-synchronization operation mechanism. Through implementation of the prototype chip and clock level simulation, it appears that the performance of a single chip multi-processor with 4 MAPLEs is close to those of modern super-scaler processors in spite of small hardware and low clock frequency.

Important Links:

Go Back