Optimizing Kernel to Kernel Communication - 2020.2 English

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

Document ID
Release Date
2020.2 English

Support for hardware accelerator pipelines that communicate through streams is one of the major advantages of FPGAs and FPGA-based SoCs and have been used in DSP and image processing applications, as well as in communication systems. As described in Streaming Data Transfers between Kernels (K2K), AXI4-Stream interfaces can be used to stream data from one kernel to another without having to use the external memory, which greatly improves the overall system latency.

Kernel ports involved in streaming are defined within the kernel, and are not addressed by the host program. There is no need to send data back to global memory before it is forwarded for processing. The connections between the kernels are directly defined during the v++ linking process as described in Specifying Streaming Connections between Compute Units.