FPGA AI Suite: Design Examples User Guide

Download PDF
ID 848957
Date 4/30/2025
Public
Document Table of Contents
Document Table of Contents x
1. FPGA AI Suite Design Examples User Guide 2. FPGA AI Suite Design Examples 3. Design Example Components 4. [PCIE] Getting Started with the FPGA AI Suite PCIe* -based Design Example 5. [PCIE] Building the FPGA AI Suite Runtime 6. [PCIE] Running the Design Example Demonstration Applications 7. [PCIE] Design Example System Architecture for the Agilex™ 7 FPGA 8. [OFS-PCIE] Getting Started with Open FPGA Stack (OFS) for PCIe* -Attach Design Examples 9. [OFS-PCIE] Design Example Components 10. [HL-NO-DDR] Getting Started with the FPGA AI Suite DDR-Free Design Example 11. [HL-NO-DDR] Running the Hostless DDR-Free Design Example 12. [HL-NO-DDR] Design Example System Architecture 13. [HL-NO-DDR] Quartus® Prime System Console 14. [HL-NO-DDR] JTAG to Avalon MM Host Register Map 15. [HL-NO-DDR] Updating MIF Files 16. [HL-JTAG] Getting Started 17. [HL-JTAG] Design Example Components 18. [SOC] FPGA AI Suite SoC Design Example Prerequisites 19. [SOC] FPGA AI Suite SoC Design Example Quick Start Tutorial 20. [SOC] FPGA AI Suite SoC Design Example Run Process 21. [SOC] FPGA AI Suite SoC Design Example Build Process 22. [SOC] FPGA AI Suite SoC Design Example Quartus® Prime System Architecture 23. [SOC] FPGA AI Suite SoC Design Example Software Components 24. [SOC] Streaming-to-Memory (S2M) Streaming Demonstration A. FPGA AI Suite Example Designs User Guide Archives B. FPGA AI Suite Example Designs User Guide Revision History
2. FPGA AI Suite Design Examples x
2.1. About the PCIe* -Attach Design Example 2.2. About the Open FPGA Stack (OFS) for PCIe* -Attach Design Examples 2.3. About the Hostless DDR-Free Design Example 2.4. About the Hostless JTAG Design Example 2.5. About the SoC Design Example
3. Design Example Components x
3.1. FPGA AI Suite Design Example Utility 3.2. Example Architecture Bitstream Files 3.3. Design Example Software Components
3.1. FPGA AI Suite Design Example Utility x
3.1.1. The dla_build_example_design.py Command 3.1.2. Listing Available FPGA AI Suite Design Examples 3.1.3. Building FPGA AI Suite Design Examples
3.1.3. Building FPGA AI Suite Design Examples x
3.1.3.1. Staging FPGA AI Suite Design Example Builds 3.1.3.2. WSL 2 FPGA AI Suite Design Example Builds
3.3. Design Example Software Components x
3.3.1. OpenVINO™ FPGA Runtime Overview 3.3.2. OpenVINO™ FPGA Runtime Plugin 3.3.3. FPGA AI Suite Runtime 3.3.4. FPGA AI Suite Custom Platform 3.3.5. Memory-Mapped Device (MMD) Driver 3.3.6. FPGA AI Suite Runtime MMD API 3.3.7. Board Support Package (BSP) Overview
3.3.7. Board Support Package (BSP) Overview x
3.3.7.1. Terasic* DE10-Agilex Development Board BSP Example 3.3.7.2. Agilex™ 7 PCIe-Attach OFS-based BSP Example
5. [PCIE] Building the FPGA AI Suite Runtime x
5.1. [PCIE] CMake Targets 5.2. [PCIE] Build Options
6. [PCIE] Running the Design Example Demonstration Applications x
6.1. [PCIE] Exporting Trained Graphs from Source Frameworks 6.2. [PCIE] Compiling Exported Graphs Through the FPGA AI Suite 6.3. [PCIE] Compiling the PCIe* -based Example Design 6.4. [PCIE] Programming the FPGA Device ( Agilex™ 7) 6.5. [PCIE] Performing Accelerated Inference with the dla_benchmark Application 6.6. [PCIE] Running the Ported OpenVINO™ Demonstration Applications
6.5. [PCIE] Performing Accelerated Inference with the dla_benchmark Application x
6.5.1. [PCIE] Inference on Image Classification Graphs 6.5.2. [PCIE] Inference on Object Detection Graphs 6.5.3. [PCIE] Additional dla_benchmark Options 6.5.4. [PCIE] The dla_benchmark Performance Metrics
6.5.2. [PCIE] Inference on Object Detection Graphs x
6.5.2.1. [PCIE] The mAP and COCO AP Metrics 6.5.2.2. [PCIE] Specifying Ground Truth 6.5.2.3. [PCIE] Example of Inference on Object Detection Graphs
6.5.4. [PCIE] The dla_benchmark Performance Metrics x
6.5.4.1. [PCIE] Interpreting System Throughput and Latency Metrics
6.6. [PCIE] Running the Ported OpenVINO™ Demonstration Applications x
6.6.1. [PCIE] Example Running the Object Detection Demonstration Application
7. [PCIE] Design Example System Architecture for the Agilex™ 7 FPGA x
7.1. [PCIE] System Overview 7.2. [PCIE] Hardware
7.2. [PCIE] Hardware x
7.2.1. [PCIE] PLL Adjustment
8. [OFS-PCIE] Getting Started with Open FPGA Stack (OFS) for PCIe* -Attach Design Examples x
8.1. [OFS-PCIE] Building the FPGA AI Suite Runtime 8.2. [OFS-PCIE] Running the Design Example Demonstration Applications
8.1. [OFS-PCIE] Building the FPGA AI Suite Runtime x
8.1.1. [OFS-PCIE] CMake Targets 8.1.2. [OFS-PCIE] Build Options
8.2. [OFS-PCIE] Running the Design Example Demonstration Applications x
8.2.1. [OFS-PCIE] Setup the OFS Environment for the FPGA Device 8.2.2. [OFS-PCIE] Exporting Trained Graphs from Source Frameworks. 8.2.3. [OFS-PCIE] Compiling Exported Graphs Through the FPGA AI Suite 8.2.4. [OFS-PCIE] Compiling the OFS for PCIe* Attach Design Example 8.2.5. [OFS-PCIE] Programming the FPGA Green Bitstream 8.2.6. [OFS-PCIE] Performing Accelerated Inference with the dla_benchmark application
8.2.6. [OFS-PCIE] Performing Accelerated Inference with the dla_benchmark application x
8.2.6.1. [OFS-PCIE] Inference on Image Classification Graphs 8.2.6.2. [OFS-PCIE] Inference on Object Detection Graphs 8.2.6.3. [OFS-PCIE] Additional dla_benchmark Options 8.2.6.4. [OFS-PCIE] The dla_benchmark Performance Metrics
8.2.6.2. [OFS-PCIE] Inference on Object Detection Graphs x
8.2.6.2.1. [OFS-PCIE] The mAP and COCO AP Metrics 8.2.6.2.2. [OFS-PCIE] Specifying Ground Truth 8.2.6.2.3. [OFS-PCIE] Example of Inference on Object Detection Graphs
8.2.6.4. [OFS-PCIE] The dla_benchmark Performance Metrics x
8.2.6.4.1. [OFS-PCIE] Interpreting System Throughput and Latency Metrics
9. [OFS-PCIE] Design Example Components x
9.1. [OFS-PCIE] Hardware Components 9.2. [OFS-PCIE] Software Components
10. [HL-NO-DDR] Getting Started with the FPGA AI Suite DDR-Free Design Example x
10.1. [HL-NO-DDR] Hardware Requirements 10.2. [HL-NO-DDR] Software Requirements
12. [HL-NO-DDR] Design Example System Architecture x
12.1. [HL-NO-DDR] System Overview 12.2. [HL-NO-DDR] Hardware
12.2. [HL-NO-DDR] Hardware x
12.2.1. [HL-NO-DDR] The Modular Scatter-Gather DMA (mSGDMA) Engines 12.2.2. [HL-NO-DDR] On-Chip Memory Modules 12.2.3. [HL-NO-DDR] Platform Designer System 12.2.4. [HL-NO-DDR] PLL Adjustment
13. [HL-NO-DDR] Quartus® Prime System Console x
13.1. [HL-NO-DDR] Quartus® Prime System Console Script Options 13.2. [HL-NO-DDR] Functionality 13.3. [HL-NO-DDR] System Reset 13.4. [HL-NO-DDR] Input Data Conversion 13.5. [HL-NO-DDR] Measuring Performance
16. [HL-JTAG] Getting Started x
16.1. [HL-JTAG] Prerequisites 16.2. [HL-JTAG] Building the FPGA AI Suite Runtime 16.3. [HL-JTAG] Building an FPGA Bitstream for the JTAG Design Examples 16.4. [HL-JTAG] Programming the FPGA Device 16.5. [HL-JTAG] Preparing Graphs for Inference with FPGA AI Suite 16.6. [HL-JTAG] Performing Inference on the Agilex™ 5 FPGA E-Series 065B Modular Development Kit 16.7. [HL-JTAG] Inference Performance Measurement 16.8. [HL-JTAG] Known Issues and Limitations
16.1. [HL-JTAG] Prerequisites x
16.1.1. [HL-JTAG] Software Requirements 16.1.2. [HL-JTAG] Hardware Requirements
17. [HL-JTAG] Design Example Components x
17.1. [HL-JTAG] Hardware Components 17.2. [HL-JTAG] Software Components
18. [SOC] FPGA AI Suite SoC Design Example Prerequisites x
18.1. [SOC] Agilex™ 7 FPGA I-Series Transceiver-SoC Development Kit Hardware Requirements
19. [SOC] FPGA AI Suite SoC Design Example Quick Start Tutorial x
19.1. [SOC] Initial Setup 19.2. [SOC] Initializing a Work Directory 19.3. [SOC] (Optional) Create an SD Card Image (.wic) 19.4. [SOC] Writing the SD Card Image (.wic) to an SD Card 19.5. [SOC] Preparing SoC FPGA Development Kits for the FPGA AI Suite SoC Design Example 19.6. [SOC] Adding Compiled Graphs (AOT files) to the SD Card 19.7. [SOC] Verifying FPGA Device Drivers 19.8. [SOC] Running the Demonstration Applications
19.3. [SOC] (Optional) Create an SD Card Image (.wic) x
19.3.1. [SOC] Installing Prerequisite Software for Building an SD Card Image 19.3.2. [SOC] Building the FPGA Bitstreams 19.3.3. [SOC] Installing HPS Disk Image Build Prerequisites 19.3.4. [SOC] (Optional) Downloading the ImageNet Categories 19.3.5. [SOC] Building the SD Card Image
19.5. [SOC] Preparing SoC FPGA Development Kits for the FPGA AI Suite SoC Design Example x
19.5.1. [SOC] Preparing the Agilex™ 5 FPGA E-Series 065B Modular Development Kit 19.5.2. [SOC] Preparing the Agilex™ 7 FPGA I-Series Transceiver-SoC Development Kit 19.5.3. [SOC] Preparing the Arria® 10 SX SoC FPGA Development Kit 19.5.4. [SOC] Configuring the SoC FPGA Development Kit UART Connection 19.5.5. [SOC] Determining the SoC FPGA Development Kit IP Address
19.5.1. [SOC] Preparing the Agilex™ 5 FPGA E-Series 065B Modular Development Kit x
19.5.1.1. [SOC] Confirming the Agilex™ 5 FPGA E-Series 065B Modular Development Kit Board Setup 19.5.1.2. [SOC] Programming the Agilex™ 5 FPGA Device with the JTAG Indirect Configuration (.jic) File 19.5.1.3. [SOC] Programming the Agilex™ 5 FPGA Device with the SRAM Object File (.sof) 19.5.1.4. [SOC] Connecting the Agilex™ 5 FPGA E-Series 065B Modular Development Kit to the Host Development System
19.5.2. [SOC] Preparing the Agilex™ 7 FPGA I-Series Transceiver-SoC Development Kit x
19.5.2.1. [SOC] Confirming Agilex™ 7 FPGA I-Series Transceiver-SoC Development Kit Board Set Up 19.5.2.2. [SOC] Programming the Agilex™ 7 FPGA Device with the JTAG Indirect Configuration (.jic) File 19.5.2.3. [SOC] Programming the Agilex™ 7 FPGA Device with the SRAM Object File (.sof) 19.5.2.4. [SOC] Connecting the Agilex™ 7 FPGA I-Series Transceiver-SoC Development Kit to the Host Development System
19.5.3. [SOC] Preparing the Arria® 10 SX SoC FPGA Development Kit x
19.5.3.1. [SOC] Confirming Arria® 10 SX SoC FPGA Development Kit Board Settings 19.5.3.2. [SOC] Connecting the Arria® 10 SX SoC FPGA Development Kit to the Host Development System
19.6. [SOC] Adding Compiled Graphs (AOT files) to the SD Card x
19.6.1. [SOC] Preparing OpenVINO™ Model Zoo 19.6.2. [SOC] Preparing a Model 19.6.3. [SOC] Compiling the Graphs 19.6.4. [SOC] Copying the Compiled Graphs to the SD card
19.8. [SOC] Running the Demonstration Applications x
19.8.1. [SOC] Running the M2M Mode Demonstration Application 19.8.2. [SOC] Running the S2M Mode Demonstration Application 19.8.3. [SOC] Troubleshooting the Demonstration Applications
20. [SOC] FPGA AI Suite SoC Design Example Run Process x
20.1. [SOC] Exporting Trained Graphs from Source Frameworks 20.2. [SOC] Compiling Exported Graphs Through the FPGA AI Suite
21. [SOC] FPGA AI Suite SoC Design Example Build Process x
21.1. [SOC] Building the Quartus® Prime Project 21.2. [SOC] Building the Bootable SD Card Image (.wic)
21.1. [SOC] Building the Quartus® Prime Project x
21.1.1. [SOC] Quartus® Prime Build Flow 21.1.2. [SOC] Build Script Options 21.1.3. [SOC] Build Directory
21.1.1. [SOC] Quartus® Prime Build Flow x
21.1.1.1. [SOC] Build Synchronization of FPGA with Software
21.1.3. [SOC] Build Directory x
21.1.3.1. [SOC] The build_stream_controller.sh Script
22. [SOC] FPGA AI Suite SoC Design Example Quartus® Prime System Architecture x
22.1. [SOC] FPGA AI Suite SoC Design Example Inference Sequence Overview 22.2. [SOC] Memory-to-Memory (M2M) Variant Design 22.3. [SOC] Streaming-to-Memory (S2M) Variant Design 22.4. [SOC] Top Level 22.5. [SOC] The SoC Design Example Platform Designer System 22.6. [SOC] Fabric EMIF Design Component 22.7. [SOC] PLL Configuration
22.2. [SOC] Memory-to-Memory (M2M) Variant Design x
22.2.1. [SOC] The mSGDMA Intel FPGA IP 22.2.2. [SOC] RAM considerations
22.3. [SOC] Streaming-to-Memory (S2M) Variant Design x
22.3.1. [SOC] Streaming Enablement for FPGA AI Suite 22.3.2. [SOC] Nios® V Subsystem 22.3.3. [SOC] Streaming System Operation 22.3.4. [SOC] Resolving Input Rate Mismatches Between the FPGA AI Suite IP and the Streaming Input 22.3.5. [SOC] The Layout Transform IP as an Application-Specific Block
22.3.3. [SOC] Streaming System Operation x
22.3.3.1. [SOC] Streaming System Buffer Management 22.3.3.2. [SOC] Streaming System Inference Job Management
22.3.5. [SOC] The Layout Transform IP as an Application-Specific Block x
22.3.5.1. [SOC] Layout Transform Considerations 22.3.5.2. [SOC] Layout Transform IP Register Map 22.3.5.3. [SOC] Layout Transform Configuration Options
22.4. [SOC] Top Level x
22.4.1. [SOC] Clock Domains
22.5. [SOC] The SoC Design Example Platform Designer System x
22.5.1. [SOC] The dla_0 Platform Designer Layer (dla.qsys) 22.5.2. [SOC] The hps_0 Platform Designer Layer (hps.qys)
23. [SOC] FPGA AI Suite SoC Design Example Software Components x
23.1. [SOC] Yocto Build and Runtime Linux Environment 23.2. [SOC] FPGA AI Suite Runtime Plugin 23.3. [SOC] Runtime Interaction with the MMD Layer 23.4. [SOC] MMD Layer Hardware Interaction Library
23.1. [SOC] Yocto Build and Runtime Linux Environment x
23.1.1. [SOC] Yocto Recipe: recipes-core/images/coredla-image.bb 23.1.2. [SOC] Yocto Recipe: recipes-bsp/u-boot/u-boot-socfpga_%.bbappend 23.1.3. [SOC] Yocto Recipe: recipes-drivers/msgdma-userio/msgdma-userio.bb 23.1.4. [SOC] Yocto Recipe: recipes-drivers/uio-devices/uio-devices.bb 23.1.5. [SOC] Yocto Recipe: recipes-kernel/linux/linux-socfpga-lts_%.bbappend 23.1.6. [SOC] Yocto Recipe: recipes-support/devmem2/devmem2_2.0.bb 23.1.7. [SOC] Yocto Recipe: wic
23.4. [SOC] MMD Layer Hardware Interaction Library x
23.4.1. [SOC] MMD Layer Hardware Interaction Library Class mmd_device 23.4.2. [SOC] MMD Layer Hardware Interaction Library Class uio_device 23.4.3. [SOC] MMD Layer Hardware Interaction Library Class dma_device
24. [SOC] Streaming-to-Memory (S2M) Streaming Demonstration x
24.1. [SOC] Nios® Subsystem 24.2. [SOC] Building the Stream Controller Module 24.3. [SOC] Building the Streaming Demonstration Applications 24.4. [SOC] Running the Streaming Demonstration
24.1. [SOC] Nios® Subsystem x
24.1.1. [SOC] Stream Controller Communication Protocol 24.1.2. [SOC] Buffer Flow in Streaming Mode using Nios® V Software Scheduler
24.1.2. [SOC] Buffer Flow in Streaming Mode using Nios® V Software Scheduler x
24.1.2.1. [SOC] Review of M2M mode 24.1.2.2. [SOC] External Streaming Mode Buffer Flow 24.1.2.3. [SOC] Nios® V Stream Controller State Machine Buffer Flow
24.4. [SOC] Running the Streaming Demonstration x
24.4.1. [SOC] The streaming_inference_app Application 24.4.2. [SOC] The image_streaming_app Application
B. FPGA AI Suite Example Designs User Guide Revision History x
B.1. FPGA AI Suite PCIe-based Design Example User Guide Document Revision History B.2. FPGA AI Suite SoC Design Example User Guide Document Revision History
1. FPGA AI Suite Design Examples User Guide
2. FPGA AI Suite Design Examples
3. Design Example Components
4. [PCIE] Getting Started with the FPGA AI Suite PCIe* -based Design Example
5. [PCIE] Building the FPGA AI Suite Runtime
6. [PCIE] Running the Design Example Demonstration Applications
7. [PCIE] Design Example System Architecture for the Agilex™ 7 FPGA
8. [OFS-PCIE] Getting Started with Open FPGA Stack (OFS) for PCIe* -Attach Design Examples
9. [OFS-PCIE] Design Example Components
10. [HL-NO-DDR] Getting Started with the FPGA AI Suite DDR-Free Design Example
11. [HL-NO-DDR] Running the Hostless DDR-Free Design Example
12. [HL-NO-DDR] Design Example System Architecture
13. [HL-NO-DDR] Quartus® Prime System Console
14. [HL-NO-DDR] JTAG to Avalon MM Host Register Map
15. [HL-NO-DDR] Updating MIF Files
16. [HL-JTAG] Getting Started
17. [HL-JTAG] Design Example Components
18. [SOC] FPGA AI Suite SoC Design Example Prerequisites
19. [SOC] FPGA AI Suite SoC Design Example Quick Start Tutorial
20. [SOC] FPGA AI Suite SoC Design Example Run Process
21. [SOC] FPGA AI Suite SoC Design Example Build Process
22. [SOC] FPGA AI Suite SoC Design Example Quartus® Prime System Architecture
23. [SOC] FPGA AI Suite SoC Design Example Software Components
24. [SOC] Streaming-to-Memory (S2M) Streaming Demonstration
A. FPGA AI Suite Example Designs User Guide Archives
B. FPGA AI Suite Example Designs User Guide Revision History

8.2.6.1. [OFS-PCIE] Inference on Image Classification Graphs

The demonstration application requires the OpenVINO™ device flag to be either HETERO:FPGA,CPU for heterogeneous execution or HETERO:FPGA for FPGA-only execution.

The dla_benchmark demonstration application runs five inference requests (batches) in parallel on the FPGA, by default, to achieve optimal system performance. To measure steady state performance, you should run multiple batches (using the niter flag) because the first iteration is significantly slower with FPGA devices.

The dla_benchmark demonstration application also supports multiple graphs in the same execution. You can place more than one graphs or compiled graphs as input, separated by commas.

Each graph can have either a different input dataset or use a commonly shared dataset among all graphs. Each graph requires an individual ground_truth_file file, separated by commas. If some ground_truth_file files are missing, the dla_benchmark continues to run and ignore the missing ones.

When multi-graph is enabled, the -niter flag represents the number of iterations for each graph, so the total number of iterations becomes -niter × number of graphs.

The dla_benchmark demonstration application switches graphs after submitting -nireq requests. The request queue holds the number of requests up to -nireq × number of graphs. This limit is constrained by the DMA CSR descriptor queue size (64 per hardware instance).

The board you use determines the number of instances that you can compile the FPGA AI Suite hardware for. For the Agilex™ 7 FPGA I-Series Development Kit and Intel® FPGA SmartNIC N6001-PL Platform, you can compile up to four instances with the same architecture on all instances. Some large architecture might not fit on the board for four instances, such as AGX7_Performance_Giant.

Each instance accesses one of the DDR banks on the board and executes the graph independently. This optimization enables multiple batches to run in parallel, limited by the number of DDR banks available. Each inference request created by the demonstration application is assigned to one of the instances in the FPGA plugin.

To enable memory-mapped device (MMD) debug messages when you run the dla_benchmark demonstration application. set the MMD_ENABLE_DEBUG environment variable as follows: 
MMD_ENABLE_DEBUG=1
Also, you can test full DDR write and read back functionality when the dla_benchmark demonstration application runs by setting the COREDLA_RUNTIME_MEMORY_TEST environment variable as follows: 
COREDLA_RUNTIME_MEMORY_TEST=1

To ensure that batches are evenly distributed between the instances, you must choose an inference request batch size that is a multiple of the number of FPGA AI Suite instances. For example, with two instances, specify the batch size as six (instead of the OpenVINO™ default of five) to ensure that the experiment meets this requirement.

The example usage that follows has the following assumptions:
  • A Model Optimizer IR .xml file is in demo/models/public/resnet-50-tf/FP32/
  • An image set is in demo/sample_images/
  • The board is programmed with a bitstream that corresponds to AGX7_Performance.arch 
binxml=$COREDLA_ROOT/demo/models/public/resnet-50-tf/FP32

imgdir=$COREDLA_ROOT/demo/sample_images

cd $COREDLA_ROOT/runtime/build_Release

./dla_benchmark/dla_benchmark \
   -b=1 \
   -m $binxml/resnet-50-tf.xml \
   -d=HETERO:FPGA,CPU \
   -i $imgdir \
   -niter=4 \
   -plugins ./plugins.xml \
   -arch_file $COREDLA_ROOT/example_architectures/AGX7_Performance.arch \
   -api=async \
   -groundtruth_loc $imgdir/TF_ground_truth.txt \
   -perf_est \
   -nireq=8 \
   -bgr
Level Two Title