Waveflow Data Schemas

Waveflow’s data schemas provide a general method for describing how data is stored in and communicated between different hardware modules. Since Waveflow is Python-based, each data structure’s schema is represented by a DataSchema class. The DataSchema abstraction provides:

  • A detailed specification of data types and fields, including bitwidths—fully compatible with the general precision types supported by Vitis HLS.
  • Methods for representing and manipulating data structures in Python. When possible, these types are mapped to NumPy arrays to enable efficient, vectorized processing.
  • Methods for automatically generating Vitis HLS-compatible C++ header files. The generated headers define C++ structs matching the schema fields, along with templated serialization/deserialization routines for arbitrary bit widths. Supported interfaces include general arrays, HLS streams, and AXI4 streams.

In this way, data schemas offer a consistent and reliable mapping between Python models and Vitis HLS implementations. The translation is automatic, eliminating manual boilerplate and error-prone hand-written packing/unpacking. As we’ll see, Data Schemas are also central for specifying strongly-typed transactional interfaces between modules.

This section splits in two:

For synthesis-pipeline details beyond schema headers, see Synthesis.

Table of contents

  • Python - The Python data-schema model — fields, lists, arrays, unions, and the fixed-point / complex element types — the single source of truth for layout and values.
  • HLS - The synthesizable C++ a schema generates — the struct codegen, and single-schema serialization (pack/read/write one value over each interface).
This site uses Just the Docs, a documentation theme for Jekyll.