Cascade Feature - 2024.1 English

Vitis Libraries

Release Date
2024.1 English

The mixed radix FFT cascade feature is configured using the TP_CASC_LEN template parameter. This determines the number of kernels over which the FFT function is split. To be clear, this feature does not use the cascade ports of kernels to convey any data. IObuffers are used to convey data from one kernel to the next in the chain. The term cascade is used simply in the sense that the function is split into a series of operations which are executed by a series of kernels, each on a separate tile. The FFT function is only split at stage boundaries, so the TP_CASC_LEN value cannot exceed the number of stages for that FFT.

Use of the cascade feature to split the FFT operation over multiple tiles will improve the throughput, because each tile in question will have fewer ranks of processing to execute and so it is ready for a new frame to process earlier.