An Efficient Self-Transposing Memory Structure for 32-bit Video Processors
Mahdi Nazm Bojnordi, Naser Sedaghati-Mokhtari, Omid Fatemi, Mahmoud Reza Hashemi · 2006
Many 2D data processing applications can be simplified and represented by use of 1D operations. Such tools, however, require applying both vertical and horizontal operations to the data blocks. The data transposing units is preferred to be used by the designers rather than applying individual operations for horizontal and vertical directions. Hence, designing a cost efficient and extendible transposing memory is a key issue for these applications. This paper proposes an efficient management strategy for using the SRAM modules in order to make a self-transposing memory architecture (STMA). In addition to its lower cost compared to flip-flop based buffers, the proposed architecture is more than 29% faster than usual SRAM based memory units. Simulations indicate that using the STMA in the H.264/AVC deblocking filter results in 60% speed improvement