The authors propose parallel implementation of prime-factor discrete cosine transform (DCT) on the orthogonal-multiprocessor (OMP) architecture when the transform size N can be decomposed into two mutually prime members N=N/sub 1/N/sub 2/. The implementation shows that the existing prime-factor DCT algorithm can be mapped easily on the OMP architecture without modification. The proposed algorithms include input index mapping, summation, scaling, adjust rotation, and output index mapping. The time complexity of the algorithm is O(N/sub 1/+N/sub 2/) on a J processor OMP, where J is the maximum dimension of Winograd-Hartley scaling matrices.
展开▼
机译:当变换大小N可以分解为两个互质的成员N = N / sub 1 / N / sub 2 /时,作者建议在正交多处理器(OMP)体系结构上并行执行质数离散余弦变换(DCT)。该实现表明,可以将现有的素数DCT算法轻松地映射到OMP体系结构,而无需进行修改。所提出的算法包括输入索引映射,求和,缩放,调整旋转和输出索引映射。该算法的时间复杂度在J处理器OMP上为O(N / sub 1 / + N / sub 2 /),其中J是Winograd-Hartley缩放矩阵的最大维。
展开▼