Improving the efficiency of parallel computing in dataflow systems
Zhabin V.I., Zhabina V.V.
In respect of fine-grained algorithms implementation there are considered parallel systems on the base of processors which perform operations using fixed and floating point arithmetic as well as redundant numerical systems for bit-by-bit data input and output. It is shown that such computing modules enable to partly execute simultaneously sequences of operations as well as to simplify respective system communication structure. The use of floating point arithmetic for operation execution enables to extend the range of number presentation as well as to provide the asynchronous mode for data transfer giving an opportunity to make calculations faster.
