20-aiecompiler-features

AI Engine Development See Vitis™ Development Environment on xilinx.com See Vitis™ AI Development Environment on xilinx.com

AI Engine Compiler Features

Version: Vitis 2024.1

Introduction

Introduction

The aiecompiler is evolving and therefore it may be difficult for you to keep up with new features. This tutorial shares a variety of features that are useful for AI Engine / AI Engine-ML (AIE-ML) programming to create more visible and efficient code compared to early versions of the compiler.

IMPORTANT: Before beginning the tutorial, make sure that you have installed the AMD Vitis™ Unified Software Platform 2024.1.

Data generation for this tutorial requires Python 3. The following packages are required:

• os
• sys
• numpy

All these designs will compile for AI Engine architecture (VCK190 production board using ARCH=aie in make command line) and the AI Engine ML architecture (VEK280 board using ARCH=aie-ml in make command line)

Objectives

After completing this tutorial, you will be able to:

• Work with multirate design that allow the compiler to handle frame length mismatch between consecutive kernels in a kernel chain.
• Send output data to different other kernels for stream-based and buffer-based I/O.
• Conditionally instantiate graph objects.

This tutorial is based on simple data passthrough to avoid another level of complexity.

Tutorial Sections

Section Link	Description
Conditional Objects	Shows how to conditionally instantiate graph objects
Multirate	Explains how to handle mismatching frame length in between kernels
Multicast	Provides examples on how to manipulate stream/buffer multicasting with multirate

Support

GitHub issues will be used for tracking requests and bugs. For questions, go to support.xilinx.com.

_{Copyright © 2024 Advanced Micro Devices, Inc.}

^{Terms and Conditions}