Dmitry Denisenko, Altera Corporation
Fast Fourier transform (FFT) is the backbone of signal processing applications. For a long time now, FPGA vendors have been providing well-tuned FFT libraries to process data sets that fit in FPGA on-chip memory. But what do you do if your data set is too large? To solve this problem, the FPGA designer must now make multiple intertwined design decisions, such as considering on-chip FFT core configuration options, how many to include, how they connect and access external memory, synchronization among multiple cores, and many others. Exploring all such design decisions to create the perfect combination for the product at hand while coding in HDL is just too time-consuming and can actually leave performance on the table. With a higher level programming language such as OpenCL, however, system design exploration can be done in days.