HoloHub Developer Reference

In this guide we aim to document advanced workflows to aid developers in working with HoloHub, and to support advanced project use cases.

user@ubuntu-20-04:/media/data/github/holohub$ ./holohub build-container --verbose
Build (HOLOHUB_ROOT:/media/data/github/holohub)...
Build (gpu_type_type:dgpu)...
Build (base_img:nvcr.io/nvidia/clara-holoscan/holoscan:v0.6.0-dgpu)...
Build (docker_file_path:/media/data/github/holohub/Dockerfile)...
Build (img:holohub:ngc-v0.6.0-dgpu)...
....

Advanced Launch Options (Container)

View All Options

Run the command below to view all available launch options in the holohub script:

$ ./holohub run-container --help

Build and Launch a Local Holoscan SDK Container

To use a HoloHub container image built with a local Holoscan SDK container:

$ ./holohub run-container --img holohub:local-sdk-latest --local-sdk-root <path_to_holoscan_sdk>

where <path_to_holoscan_sdk> is the path to the Holoscan SDK root directory containing the build directory. Please refer to the Holoscan SDK Developer Guide for more details on how to build the Holoscan SDK from source.

In the container, to verify the build directory (with a python build) is mounted correctly, run the following command:

$ python -c "import holoscan; print(holoscan.__file__)"

The output should be something like:

/workspace/holoscan-sdk/build-x86_64/python/lib/holoscan/__init__.py

If Python supported is not enabled, /workspace/holoscan-sdk can be manually inspected to confirm the mount. The directory should contain a non-empty build-<arch>-<gpu_type> or install-<arch>-<gpu_type> directory.

Launch a Named HoloHub Container

To launch custom HoloHub container with fully qualified name, e.g. "holohub:ngc-sdk-sample-app"

$ ./holohub run-container --img holohub:ngc-sdk-sample-app --no-docker-build

Forward X11 Graphics Over SSH

  ./holohub run-container --ssh-x11

Support Nsight Systems profiling in the HoloHub Container

  ./holohub run-container --nsys-profile

Advanced Options for Building Applications

View Build Options

./holohub build --help

Local SDK Path

If you have an installation of the Holoscan SDK which is not in a standard path, you may want to provide the root directory of your Holoscan SDK installation.

  ./holohub build --configure-args="-Dholoscan_DIR=/path/to/holoscan/install/lib/cmake/holoscan"

Building a Specific Application

By default HoloHub builds all the sample applications that are maintained with the SDK. You can build specific applications by the name of the directory.

  ./holohub build <application>

For example:

  ./holohub build endoscopy_tool_tracking

Note that CMake will build the application in the directory specified. If there are multiple languages, the script will attempt to build all of them.

Building application or operator manually

If you prefer to build applications and operator manually you can follow the steps below.

# Export cuda (in case it's not already in the path)
export PATH=$PATH:/usr/local/cuda/bin

# Configure HoloHub with CMake
cmake -S <path_to_holohub_source>            # Source directory
      -B build                               # Build directory
      -DPython3_EXECUTABLE=/usr/bin/python3  # Specifies the python executable for CMake to find the correct version
      -DHOLOHUB_DATA_DIR=$(pwd)/data         # Specifies the data directory
      -DAPP_<name_of_the_application>=1      # Specified the application to build


# Build the application(s)
cmake --build build

Additional Build Notes

While not all applications requires building HoloHub, the current build system automatically manages dependencies (applications/operators) and also downloads and converts datasets at build time.

You can refer to the README of each application/operator if you prefer to build/run them manually.

The ./holohub script creates a data subdirectory to store the downloaded HoloHub data. This directory is noted HOLOHUB_DATA_DIR/holohub_data_dir in the documentation, READMEs and metadata files.

Advanced Options for Running Applications

Pass additional arguments to the application command

  ./holohub run endoscopy_tool_tracking --language=python --run-args='-r visualizer'

Profile using Nsight Systems

For example, to profile endoscopy_tool_tracking using Nsight Systems:

First config the app's replay count to 10 frames:

--- a/applications/endoscopy_tool_tracking/python/endoscopy_tool_tracking.yaml
+++ b/applications/endoscopy_tool_tracking/python/endoscopy_tool_tracking.yaml
@@ -81,7 +81,7 @@ replayer:
   frame_rate: 0   # as specified in timestamps
   repeat: true    # default: false
   realtime: true  # default: true
-  count: 0        # default: 0 (no frame count restriction)
+  count: 10

Then run the app with --nsys-profile option:

  ./holohub run endoscopy_tool_tracking --language=python --nsys-profile

This will create a Nsight Systems report file in the application working directory. Information on the generated report file is printed on the end of the application log:

Generating '/tmp/nsys-report-bcd8.qdstrm'
[1/1] [========================100%] report8.nsys-rep
Generated:
    /workspace/holohub/build/endoscopy_tool_tracking/report8.nsys-rep

This file can be loaded and visualized with the Nsight Systems UI application:

  nsys-ui /workspace/holohub/build/endoscopy_tool_tracking/report8.nsys-rep

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!