Enabling Host-to-Kernel Dataflow - 2020.2 English

Vitis Unified Software Platform Documentation: Application Acceleration Development (UG1393)

Document ID
UG1393
Release Date
2021-03-22
Version
2020.2 English

If a kernel is capable of accepting more data while it is still operating on data from the previous transactions, XRT can send the next batch of data. The kernel then works on multiple data sets in parallel at different stages of the algorithm, thus improving performance. To support host-to-kernel dataflow, the kernel has to implement the ap_ctrl_chain protocol using the pragma HLS interface for the function return:

void kernel_name( int *inputs,
                  ...         )// Other input or Output ports
{
#pragma HLS INTERFACE  .....   // Other interface pragmas
#pragma HLS INTERFACE ap_ctrl_chain port=return bundle=control
Important: To take advantage of the host-to-kernel dataflow, the kernel must also be written to process data in stages, such as pipelined at the loop-level as discussed in Loop Pipelining, or pipelined at the task-level as discussed in Dataflow Optimization.