Design and implementation of matrix transpose based on SMP system

The paper studies the processing method of matrix transpose in symmetric multiprocessor (SMP) parallel processing system. It proposes a two-dimensional equilibrium interleaving matrix transposition algorithm which is suitable for SMP systems by analysing two factors that influence the access efficiency of DDR. The total number of active pages in processing of accessing matrix is reduced by the processing algorithm through dividing the matrix into many small matrixes which also saves a lot of expenditure of time. It is made according with the burst character of DDR by adopting the way of interleaving storage of rows and columns in pages which increases the efficiency of accessing pages´ data. Test results show that the rows and columns have almost the same access bandwidth and two-dimensional average bandwidth can reach 86.7% of the theoretical bandwidth. The algorithm can solve the problem of matrix transpose in synthetic aperture radar system.