1. Launch Momentics and create a new workspace.

2. Create a new project based on the ZCU102 BSP 

   • File->Import

     Open

     

     
     
     

   • Choose QNX->QNX Source Package and BSP. Click Next

     Open

     

   • Browse for the ZCU102 BSP ZIP Archive

     • <QNX SDP install path>\bsp\BSP_xilinx-zynq-zcu102_br-710_be-710_SVN964557_JBN16.zip

     • Click Open, then Next Twice

     • Open

       
       Open

       

     • Give the project a suitable name

     • Click Finish

   • You should now have a new Project open in the Momentics IDE similar to the following:

     Open

     

The following steps are required for both Vitis and Vivado flows.

1. Add VAI libraries
   Several libraries need to be installed in the root file system. Add the following lines to the build file after the .script section. If the Vitis-AI Library layer is not needed, the lines after # Vitis-AI Library below can be excluded to save space, or only certain models can be included.

   ########################################################################### ## Vitis-AI ########################################################################### # Utilities & Drivers /bin/show_dpu=show_dpu # Vitis-AI libs /usr/lib/libjson-c.so=libjson-c.so /usr/lib/libprotobuf.so=libprotobuf.so /usr/lib/libprotoc.so=libprotoc.so /usr/lib/libc++.so=libc++.so /usr/lib/libiconv.so=libiconv.so /usr/lib/libintl.so=libintl.so /usr/lib/libboost_system.so=libboost_system.so /usr/lib/libboost_atomic.so=libboost_atomic.so /usr/lib/libboost_filesystem.so=libboost_filesystem.so /usr/lib/libglog.so=libglog.so /usr/lib/libgflags.so=libgflags.so /lib/libcatalog.so=libcatalog.so /lib/libunilog.so=libunilog.so /lib/libtarget_factory.so=libtarget_factory.so /lib/libxir.so=libxir.so /lib/libvart.so=libvart.so /lib/libvart-dpu-runner.so=libvart-dpu-runner.so # Vitis-AI Library libs /lib/libvitis_ai_library-extra_ops.so=libvitis_ai_library-extra_ops.so /lib/libvitis_ai_library-model_config.so=libvitis_ai_library-model_config.so /lib/libvitis_ai_library-math.so=libvitis_ai_library-math.so /lib/libvitis_ai_library-runner_helper.so=libvitis_ai_library-runner_helper.so /lib/libvitis_ai_library-dpu_task.so=libvitis_ai_library-dpu_task.so /lib/libvitis_ai_library-cpu_task.so=libvitis_ai_library-cpu_task.so /lib/libvitis_ai_library-graph_runner.so=libvitis_ai_library-graph_runner.so /lib/libvitis_ai_library-xmodel_jit.so=libvitis_ai_library-xmodel_jit.so /lib/libvitis_ai_library-xmodel_image.so=libvitis_ai_library-xmodel_image.so /lib/libvitis_ai_library-xnnpp.so=libvitis_ai_library-xnnpp.so # Vitis-AI Library model-specific libraries /lib/libvitis_ai_library-classification.so=libvitis_ai_library-classification.so /lib/libvitis_ai_library-tfssd.so=libvitis_ai_library-tfssd.so /lib/libvitis_ai_library-facedetect.so=libvitis_ai_library-facedetect.so /lib/libvitis_ai_library-facefeature.so=libvitis_ai_library-facefeature.so /lib/libvitis_ai_library-lanedetect.so=libvitis_ai_library-lanedetect.so /lib/libvitis_ai_library-yolov2.so=libvitis_ai_library-yolov2.so /lib/libvitis_ai_library-yolov3.so=libvitis_ai_library-yolov3.so /lib/libvitis_ai_library-RGBDsegmentation.so=libvitis_ai_library-RGBDsegmentation.so /lib/libvitis_ai_library-facelandmark.so=libvitis_ai_library-facelandmark.so /lib/libvitis_ai_library-facequality5pt.so=libvitis_ai_library-facequality5pt.so /lib/libvitis_ai_library-ssd.so=libvitis_ai_library-ssd.so /lib/libvitis_ai_library-segmentation.so=libvitis_ai_library-segmentation.so /lib/libvitis_ai_library-covid19segmentation.so=libvitis_ai_library-covid19segmentation.so /lib/libvitis_ai_library-3Dsegmentation.so=libvitis_ai_library-3Dsegmentation.so /lib/libvitis_ai_library-refinedet.so=libvitis_ai_library-refinedet.so /lib/libvitis_ai_library-openpose.so=libvitis_ai_library-openpose.so /lib/libvitis_ai_library-hourglass.so=libvitis_ai_library-hourglass.so /lib/libvitis_ai_library-posedetect.so=libvitis_ai_library-posedetect.so /lib/libvitis_ai_library-reid.so=libvitis_ai_library-reid.so /lib/libvitis_ai_library-rcan.so=libvitis_ai_library-rcan.so /lib/libvitis_ai_library-reidtracker.so=libvitis_ai_library-reidtracker.so /lib/libvitis_ai_library-multitask.so=libvitis_ai_library-multitask.so /lib/libvitis_ai_library-multitaskv3.so=libvitis_ai_library-multitaskv3.so /lib/libvitis_ai_library-platedetect.so=libvitis_ai_library-platedetect.so /lib/libvitis_ai_library-platenum.so=libvitis_ai_library-platenum.so /lib/libvitis_ai_library-platerecog.so=libvitis_ai_library-platerecog.so /lib/libvitis_ai_library-carplaterecog.so=libvitis_ai_library-carplaterecog.so /lib/libvitis_ai_library-medicalsegmentation.so=libvitis_ai_library-medicalsegmentation.so /lib/libvitis_ai_library-medicaldetection.so=libvitis_ai_library-medicaldetection.so /lib/libvitis_ai_library-facerecog.so=libvitis_ai_library-facerecog.so /lib/libvitis_ai_library-facedetectrecog.so=libvitis_ai_library-facedetectrecog.so /lib/libvitis_ai_library-pointpillars.so=libvitis_ai_library-pointpillars.so /lib/libvitis_ai_library-pointpillars_nuscenes.so=libvitis_ai_library-pointpillars_nuscenes.so /lib/libvitis_ai_library-pointpainting.so=libvitis_ai_library-pointpainting.so /lib/libvitis_ai_library-centerpoint.so=libvitis_ai_library-centerpoint.so /lib/libvitis_ai_library-retinaface.so=libvitis_ai_library-retinaface.so /lib/libvitis_ai_library-mnistclassification.so=libvitis_ai_library-mnistclassification.so /lib/libvitis_ai_library-medicalsegcell.so=libvitis_ai_library-medicalsegcell.so /lib/libvitis_ai_library-bcc.so=libvitis_ai_library-bcc.so /lib/libvitis_ai_library-pmg.so=libvitis_ai_library-pmg.so /lib/libvitis_ai_library-general1.so=libvitis_ai_library-general1.so /lib/libvitis_ai_library-multi_runner.so=libvitis_ai_library-multi_runner.so /lib/libvitis_ai_library-arflow.so=libvitis_ai_library-arflow.so /lib/libvitis_ai_library-bevdet.so=libvitis_ai_library-bevdet.so /lib/libvitis_ai_library-c2d2_lite.so=libvitis_ai_library-c2d2_lite.so /lib/libvitis_ai_library-cflownet.so=libvitis_ai_library-cflownet.so /lib/libvitis_ai_library-clocs.so=libvitis_ai_library-clocs.so /lib/libvitis_ai_library-efficientdet_d2.so=libvitis_ai_library-efficientdet_d2.so /lib/libvitis_ai_library-fairmot.so=libvitis_ai_library-fairmot.so /lib/libvitis_ai_library-fusion_cnn.so=libvitis_ai_library-fusion_cnn.so /lib/libvitis_ai_library-monodepth2.so=libvitis_ai_library-monodepth2.so /lib/libvitis_ai_library-movenet.so=libvitis_ai_library-movenet.so /lib/libvitis_ai_library-ocr.so=libvitis_ai_library-ocr.so /lib/libvitis_ai_library-ofa_yolo.so=libvitis_ai_library-ofa_yolo.so /lib/libvitis_ai_library-pmrid.so=libvitis_ai_library-pmrid.so /lib/libvitis_ai_library-polypsegmentation.so=libvitis_ai_library-polypsegmentation.so /lib/libvitis_ai_library-solo.so=libvitis_ai_library-solo.so /lib/libvitis_ai_library-textmountain.so=libvitis_ai_library-textmountain.so /lib/libvitis_ai_library-tracker.so=libvitis_ai_library-tracker.so /lib/libvitis_ai_library-ultrafast.so=libvitis_ai_library-ultrafast.so /lib/libvitis_ai_library-vehicleclassification.so=libvitis_ai_library-vehicleclassification.so /lib/libvitis_ai_library-yolov6.so=libvitis_ai_library-yolov6.so /lib/libvitis_ai_library-yolovx.so=libvitis_ai_library-yolovx.so /lib/libxmodel_postprocessor_common.so=libxmodel_postprocessor_common.so /lib/libxmodel_postprocessor_classification.so=libxmodel_postprocessor_classification.so /lib/libxmodel_postprocessor_densebox.so=libxmodel_postprocessor_densebox.so /lib/libxmodel_postprocessor_plate_number.so=libxmodel_postprocessor_plate_number.so /lib/libxmodel_postprocessor_face_recognition.so=libxmodel_postprocessor_face_recognition.so /lib/libxmodel_postprocessor_face_quality.so=libxmodel_postprocessor_face_quality.so /lib/libxmodel_postprocessor_face_landmark.so=libxmodel_postprocessor_face_landmark.so /lib/libxmodel_postprocessor_open_pose.so=libxmodel_postprocessor_open_pose.so /lib/libxmodel_postprocessor_segmentation.so=libxmodel_postprocessor_segmentation.so /lib/libxmodel_postprocessor_ssd.so=libxmodel_postprocessor_ssd.so /lib/libxmodel_postprocessor_yolov3.so=libxmodel_postprocessor_yolov3.so /lib/libxmodel_preprocessor_common.so=libxmodel_preprocessor_common.so /lib/libxmodel_preprocessor_vgg.so=libxmodel_preprocessor_vgg.so /lib/libxmodel_preprocessor_efficientnet.so=libxmodel_preprocessor_efficientnet.so

    

2. Add sample applications
   There are a few sample applications packaged with the Vitis-AI package. Add these lines to the build file after the Vitis-AI libraries.

   /usr/bin/resnet50=resnet50 /usr/bin/test_jpeg_facedetect=test_jpeg_facedetect /usr/bin/segmentation=segmentation

3. Add Python support
   Add the following to include python support. This is optional and only required for exercising the VAI python APIs.

4. Add OpenCV support
   Vitis-AI requires access to OpenCV, so add the following OpenCV libraries to the IFS.

   ################################# # OpenCV ################################# [prefix=/usr/lib] libopencv_calib3d.so libopencv_core.so libopencv_dnn.so libopencv_features2d.so libopencv_flann.so libopencv_highgui.so libopencv_imgcodecs.so libopencv_imgproc.so libopencv_ml.so libopencv_objdetect.so libopencv_photo.so libopencv_stitching.so libopencv_videoio.so libopencv_video.so #Restore default prefix [prefix=/proc/boot]

    

The following steps should be taken for the Vitis flow only.

1. Add AXI_INTC support to the QNX startup process.
   The Vitis platform used to implement the Vitis Flow reference design (2022.2 ZCU102 Vitis Base Platform) includes an AXI_INTC interrupt controller in the programmable logic to aggregate up to 32 interrupts from different compute units, including the DPU. The output of the AXI_INTC connects to the GIC through the PL->PS interrupt bus. Therefore, we need to tell the QNX startup process to enable and configure the AXI_INTC.
   
   In the .bootstrap section, modify the startup-xzynq-zcu10x command to add an argument to configure the base address (0x80020000) and cascaded GIC interrupt (121) of AXI_INTC IP in the PL.

   startup-xzynq-zcu10x -v -a0x80020000,121,0

2. Start the zocl driver at boot time.
   XRT relies on the zocl device driver (or QNX device resource manager) to manage communications and data transfer with the connected compute units, including the DPU. Therefore, we need to start up the driver and inform it what interrupts are assigned to it.
   
   In the .script section, AFTER /sbin/screen is run, and before the QSPI flash section, add the following command to start the zocl driver for XRT. The interrupts are design-specific and matches the configuration of the matches the ZCU102 base Vitis platform used in the DPU TRD which has 32 interrupts connected into the AXI_INTC IP.

   zocl -U 523:523 -i 8192,8193,8194,8195,8196,8197,8198,8199,8200,8201,8202,8203,8204,8205,8206,8207,8208,8209,8210,8211,8212,8213,8214,8215,8216,8217,8218,8219,8220,8221,8222,8223 &

    

3. Add XRT libs/drivers/apps to the filesystem.
   Several different libraries, the zocl driver and the xbutil utility need to be installed in the root file system. Add the following lines to the build file after the .script section. We add them after the # Shared libraries section and before the ## Tracelog section:
   
   

   ########################################################################## ## Xilinx XRT ########################################################################### /lib/libxrt_coreutil.so=libxrt_coreutil.so /lib/libxrt_core.so=libxrt_core.so /lib/libxrt++.so=libxrt++.so /lib/libxilinxopencl.so=libxilinxopencl.so /lib/libxdp_core.so=libxdp_core.so /lib/libxdp_appdebug_plugin.so=libxdp_appdebug_plugin.so /lib/libxdp_debug_plugin.so=libxdp_debug_plugin.so /lib/libxdp_device_offload_plugin.so=libxdp_device_offload_plugin.so /lib/libxdp_hal_api_interface_plugin.so=libxdp_hal_api_interface_plugin.so /lib/libxdp_hal_device_offload_plugin.so=libxdp_hal_device_offload_plugin.so /lib/libxdp_hal_plugin.so=libxdp_hal_plugin.so /lib/libxdp_lop_plugin.so=libxdp_lop_plugin.so /lib/libxdp_native_plugin.so=libxdp_native_plugin.so /lib/libxdp_noc_plugin.so=libxdp_noc_plugin.so /lib/libxdp_opencl_counters_plugin.so=libxdp_opencl_counters_plugin.so /lib/libxdp_opencl_trace_plugin.so=libxdp_opencl_trace_plugin.so /lib/libxdp_pl_deadlock_plugin.so=libxdp_pl_deadlock_plugin.so /lib/libxdp_power_plugin.so=libxdp_power_plugin.so /lib/libxdp_system_compiler_plugin.so=libxdp_system_compiler_plugin.so /lib/libxdp_user_plugin.so=libxdp_user_plugin.so /lib/libxdp_vart_plugin.so=libxdp_vart_plugin.so /lib/libsecpol.so=libsecpol.so /lib/libcatalog.so=libcatalog.so /usr/lib/libc++.so=libc++.so /usr/lib/libboost_filesystem.so=libboost_filesystem.so /usr/lib/libboost_system.so=libboost_system.so /usr/lib/libboost_program_options.so=libboost_program_options.so /usr/lib/libboost_atomic.so=libboost_atomic.so /bin/zocl=zocl /usr/bin/xbutil=xbutil

4. Add XLNX_VART_FIRMWARE environment variable
   Because dpu.xclbin is on the SD card, we need to set an environment variable so that VAI knows where to find it. Add the following line inside the section where /etc/profile is defined:

   export XLNX_VART_FIRMWARE=/dpu.xclbin

The following steps should be taken for the Vivado flow only.

1. Add XRT Minimal Libraries

Since the VART library is shared between the two flows, a minimal set of XRT libraries is required even for the Vivado flow. Add the following after the # Shared libraries section and before the ## Tracelog section:

2. Add the DPU Driver

First, in the section where we added the Vitis-AI libraries, add the DPU driver:

Next, in the .script section AFTER /sbin/screen is run and before the QSPI flash section, add the following command to start the DPU driver for Vivado flow.

In this example the base address and interrupt information match that of the default VAI 3.0 DPU TRD for the Vivado flow:

• DPU Base Address (-a)

  • 0x8F00_0000

• Interrupts (-i)

  • In this design, there are four interrupts from the DPU IP - one for each DPU core, and one for the Softmax core. Interrupts are connected directly from the DPU IP to the PS GIC via the upper
    PL->PS Interrupt interface (ps_ps_irq0[3:0])

  • Source/Int #/GIC #: DPU0/0x59/121, DPU1/0x5a/122, DPU2/0x5b/123, SFM/0x5c/124

Next, add some useful utilities at the end of the .build file:

To mount the FAT partition, first, add the following script to the .build file right after the .script section:

Now call the script from the main startup .script right before the call to start the main shell:

The FAT partition will be mounted at / in the QNX filesystem.

1. First, make sure you have saved zcu102.build.

2. Right click on the project name and select "Clean Project”

3. Right click on the project name and select "Build Project"

4. Open

   

   Wait for the build to complete. This process will take several minutes.

5. Once complete, the QNX-IFS (QNX Initial File System) will be available in the images folder.