Single Source Shortest Path Application - 2023.2 English

Vitis Tutorials: Hardware Acceleration (XD099)

Document ID
XD099
Release Date
2023-11-13
Version
2023.2 English

Vitis™ Hardware Acceleration Tutorials

See Vitis™ Development Environment on xilinx.com

Single Source Shortest Path Application with Vitis Libraries for Acceleration

Version: Vitis 2023.2

Design your application using the optimized AMD Vitis™ accelerated libraries. Vitis libraries offer out-of-the-box acceleration with minimal to zero-code changes to your applications, accelerating your x86 host application code calling Vitis accelerated library APIs or developing accelerators on AMD platforms calling kernels in your code. You can work at an application level and focus your core competencies on solving problems by using the libraries in commonly-used programming languages and leveraging AMD platforms as an enabler in your application.

Problem Description

In the field of transportation, many problems and algorithms are based on traffic network data. We assume the city traffic networks as weighted direct graph composed with nodes and links. The following is an example diagram:

1
1
2
2
3
3
8
8
5
5
9
9
7
7
4
4
6
6
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
weights
Text is not SVG - cannot display

This example shows how to calculate the distance from the source point to other points. The first node is taken as the source vertice.Weights may have a variety of formats, including cost and traveltime.

  • Calculate the traveltime according to the BPR function

    LinkTravelTime
    LinkTravel...
    =
    =
    FreeFlowTime
    FreeFl...
    The Bureau of Public Roads (BPR) function, relating the travel time to
    the flow. 
    The Bureau of Public Roads (BPR) func...
    x
    x
    power
    power
    Capacity
    Capacity
    Flow
    Flow
    x
    x
    b
    b
    1 +
    1 +
    (
    (
    )
    )
    ______
    ______
    (
    (
    )
    )
    Text is not SVG - cannot display

  • Calculate the weight average distance

    weights
    weights
    =
    =
    0.6*Cost
    0.6*Co...
    You can calculate the weighted average in other functions.
    This is only for reference.
     
    You can calculate the weighted...
    +
    +
    0.4*Traveltime
    0.4*Tr...
    Text is not SVG - cannot display

  • Compute the Shortest Path

    Bellman-Ford algorithm equiped with a First-In-First-Out queue
    Bellman-Ford algorithm equiped wi...
    Enqueue
    Enqueue
    Dequeue
    Dequeue
    Text is not SVG - cannot display

About This Tutorial

This tutorial introduces an application developed by a Single Source Shortest Path (SSSP) kernel krnls_sssp based on the Vitis Graph Library L2 APIs. The L2 SSSP API provides an HLS function that can be directly built into the Vitis compute-unit (OpenCL kernel). The application also includes other kernels, krnls_wa and krnls_search, as well as the host-kernel interaction with OpenCL. The krnls_wa kernel prepares the weight average traffic weights data. The krnls_search kernel retrieves the results. The Vitis Library takes care of the heavy algorithm lifting so that you can focus on other kernels and host codes.

Note: All the steps in this tutorial except those needed to view system diagram and optional part shows the Vitis GUI flow use the command-line interface.

Global

 Memory
Global...
krnls_wa
krnls_wa
krnls_sssp
krnls_sssp
krnls_search
krnls_search
Input
Data
Input...
Input
Data
Input...
Output
Data
Output...
Result
Data
Result...
AXI stream
connection
AXI stream...
Result
Data
Result...
AXI stream
connection
AXI stream...
Text is not SVG - cannot display

The AMD Alveo™ Data Center accelerator card U50 is the target platform for this tutorial. The following tasks are described in this tutorial:

  • Using the Vitis Graph Library L2 design the Single Source Shortest Path kernel: krnls_sssp

  • Designing other kernels with C++: krnls_wa, krnls_search

  • Programming the host with OpenCL APIs

  • Writing the Makefile

  • Preparing the dataset for running the application

  • Using Vitis hardware to run the application

Note: This is not a detailed tools usage guideline, so only basic ideas and steps are shown. Refer to other documents for details.

Tutorial Structure

This tutorial is divided into three sections.

  • Section 1 - Workflow

    • UsIng the SSSP kernel based on Vitis Graph Library L2

    • Designing other kernels

    • Programming the host

    • Writing the Makefile

  • Section 2 - Setup_Environment

    • This section provides information on setting up the environment.

  • Section 3 - Application

    • Part 1: Preparing data, including the offset, indices, and weights, in the CSR format

    • Part 2: Compiling codes and running the application

    • Part 3: Profiling with the Vitis tool

    • Part 4: Creating a project with the Vitis GUI and debugging. Skip this part if you do not want to use GUI.

Summary

This tutorial demonstrates how to expand your code with PL Vitis Libraries L2 APIs and describes the development flow based on the Vitis software platform. The purpose of this application in this example is to find the shortest path from the source node to others using city traffic networks as the direct weight graph. You can also develop applications with the other libraries using the same flow.

Reference

Documents on Vitis Libraries Docs.

Copyright © 2020–2023 Advanced Micro Devices, Inc

Terms and Conditions