This AI Engine kernel implements the DFT-7 using a vector-matrix approach based on mac8() intrinsics to compute two [1x1] x [1x4] OPs in parallel in a single cycle. Two compute tiles and 1 combiner tile are required. The first compute tile is 100% efficient whereas the second compute tile is 75% efficient. Input samples for two separate transforms arrive over two streams, one transform on each stream, over a period of seven cycles. The output samples are produced alternately for each transform on the two output streams over the same period. The graph for this AI Engine kernel is shown below.