METHOD OF ACCELERATOR NDP for solving large systems of linear equations

摘要

Method organization arithmetic accelerator to solve large systems of linear equations comprising a shared memory available to the primary and one or more secondary processors, receiving a plurality of coefficients associated with a set of linear equations in the shared memory and their division into coefficient blocks using one or more primary processor , the identification of available secondary processor for processing selected block of coefficients, transmitting it from the shared memory block in the secondary processor local memory Seq.Id.No. blowing treatment to produce an intermediate result and transmitting it to the shared memory, characterized in that the accesses to shared memory block of one or more tertiary or quaternary processors selected from an arbitrary plurality of heterogeneous processors, identifying a free primary processor and partitioned intermediate result to groups produce indexing and write an intermediate result values ​​in each group in a block of shared memory, identify the processor and free tertiary produce rankings ind eksov and one of the three successive indices selected from a plurality of indexes produce fast discrete Fourier transform module recorded the conversion results in a block of shared memory, identify free quaternary processor sequentially sorted first index and produce fast discrete Fourier transformation of two other indices, multiplied pointwise values ​​for these two indices on the Fourier transformation unit Toeplitz matrix indices for these produce discrete inverse fast converted