This paper presents a systematic methodology based on the algebraic theory of signal processing to classify and derive fast algorithms for linear transforms. Instead of manipulating the entries of transform matrices, our approach derives the algorithms by stepwise decomposition of the associated signal models, or polynomial algebras. This decomposition is based on two generic methods or algebraic principles that generalize the wellknown Cooley-Tukey FFT and make the algorithms' derivations concise and transparent. Application to the 16 discrete cosine and sine transforms yields a large class of fast algorithms, many of which have not been found before.

The Cell BE is a multicore processor with eight vector accelerators (called SPEs) that implement explicit cache management through direct memory access engines. While the Cell has an impressive floating point peak performance, programming and optimizing for it is difficult as it requires explicit memory management, multithreading, streaming, and vectorization. We address this problem for the discrete Fourier transform (DFT) by extending Spiral, a program generation system, to automatically generate highly optimized implementations for the Cell. The extensions include multi-SPE parallelization and explicit memory streaming, both performed at a high abstraction level using rewriting systems operating on Spiral's internal domain-specific language. Further, we support latency and throughput optimizations, single and double precision, and different data formats. The performance of Spiral's computer generated code is comparable with and sometimes better than existing DFT implementations, where available.

No abstract

No abstract

No abstract