Vitis HLS implements channels between the tasks as either PIPO or FIFO buffers, depending on the user's choice:
- For scalars, Vitis HLS will automatically infer FIFOs as the channel type.
- If the parameter (of a producer or consumer) is an array, the user has a choice of
implementing the channel as a PIPO or a FIFO based on the following
considerations:
- If the data is always accessed in sequential order, the user can choose to implement this memory channel as PIPO/FIFO. Choosing PIPOs comes with the advantage that PIPOs can never deadlock but they require more memory to use. Choosing FIFOs offers the advantage of lesser memory requirements but this comes with the risk of deadlock if the FIFO sizes are not correct.
-
If the data is accessed in an arbitrary manner, the memory channel must be implemented as a PIPO (with a default size that is twice the size of the original array).
Tip: A PIPO ensures that the channel always has the capacity to hold all samples produced in one iteration, without a loss.
Specifying the size of the FIFO channels overrides the default value that is computed by the tool to attempt to optimize the throughput. If any function in the design can produce or consume samples at a greater rate than the specified size of the FIFO, the FIFOs might become empty (or full). In this case, the design halts operation, because it is unable to read (or write). This might lead to a stalled, deadlock state.
When setting the depth of the FIFOs, AMD recommends initially setting the depth as the maximum number of data values transferred (for example, the size of the array passed between tasks), confirming the design passes C/RTL co-simulation, and then reducing the size of the FIFOs and confirming C/RTL co-simulation still completes without issues. If RTL co-simulation fails, the size of the FIFO is likely too small to prevent stalling or a deadlock situation. The Vitis HLS GUI now supports an automatic way of determining the right FIFO size to use. Additionally, the Vitis HLS IDE can display a histogram of the size of each FIFO/PIPO buffer over time, after RTL co-simulation has been run. This can be useful to help determine the best depth for each buffer.