Читать книгу Multi-Processor System-on-Chip 2 - Liliana Andrade - Страница 25

The associated dataflow processing graph is depicted in Figure 1.7, with the notation foo[bar] abbreviated as foobar.

Each multiply-accumulate (MAC) in Figure 1.7 operates on input/output vector length of K elements that results in the total M ² K scalar MAC operations, 2M ² K loads and MK stores. This is the theoretical minimum number of operations. The MAC result is stored in an indexed accumulator register (ACC) and is used in the next MAC of the same index. For readers not familiar with dataflow diagrams, a matrix representation of GFDM filtering is shown in Figure 1.8, and the experience drawn from matrix multiplication implementations can also be used here.

Figure 1.7. GFDM processing dataflow diagram