GstCUDA - Evaluating GstCUDA

GstCUDA
Features and Limitations
Supported Platforms
Evaluating GstCUDA
Building and Installation Guide
Project Structure
GstCUDA Framework
libGstCUDA API GstCUDABaseFilter GstCUDABaseMISO GstCUDABaseSIMO GstCUDABaseMIMO
GstCUDA Quick Prototyping Elements
cudafilter cudamux cudademux cudamimo
GstCUDA Algorithm Developer’s Guide
GstCUDA Add-Ons
cudadebayer
Examples
cudafilter cudafilter: NVMM direct mapping cudafilter: Unified memory allocator cudamux cudamux: NVMM direct mapping cudamux: Unified memory allocator cudadebayer opencv opencvwarp opencvfilter
Performance Profiling
Contact Us

This page serves as a guide to test the GstCUDA evaluation binary provided by RidgeRun.

Requesting the Evaluation Binary

In order to request an evaluation binary for a specific architecture, please contact us providing the following information:

• Platform (i.e.: TX1, TX2, x86)
• gst-launch-1.0 --gst-version
• uname -a
• Jetpack and L4T version (Linux distribution version for x86)

Evaluation Version Feature

The GstCUDA evaluation version contains all the features of the full version, with the exception that it is limited to process a maximum of 1800 frames. Also, it is provided in binary form, so you won't have access to the source code.

The 1800 frames maximum processing limitation allows to test GstCUDA for 30 seconds on a 60 fps pipeline, or for one minute on a 30 fps pipeline.

The following table summarizes the features available in both the professional and evaluation version of the element.

Table 1. Features provided in the evaluation version

Feature	Professional	Evaluation
cudafilter Element	Y	Y
cudaexamplefilter Element	Y	Y
CUDA algorithm examples Library	Y	Y
Unlimited Processing Time	Y	N (1)
Source Code	Y	N

(1) The evaluation version will limit the processing to a maximum of 1800 frames.

Installation and testing

Dependencies

First, make sure the dependencies are fulfilled:

sudo apt-get install \
libgstreamer1.0-dev \
libgstreamer-plugins-base1.0-dev \
gstreamer1.0-plugins-base \
libgstreamer-plugins-bad1.0-dev \
gstreamer1.0-plugins-bad \
libegl1-mesa-dev

Also, be sure that [CUDA L4T package] is successfully installed.

Eval Binary structure

The provided GstCUDA eval version tarball must have the following structure:

.
├── examples
│   ├── algorithm.mk
│   ├── cudafilter_algorithms
│   │   ├── gray-scale-filter
│   │   │   ├── gray-scale-filter.cu
│   │   │   ├── gray-scale-filter.so
│   │   │   ├── Makefile
│   │   │   ├── README.md
│   │   │   └── test-pipelines.txt
│   │   ├── median-filter
│   │   │   ├── Makefile
│   │   │   ├── median-filter.cu
│   │   │   ├── median-filter.so
│   │   │   └── README.md
│   │   ├── memcpy
│   │   │   ├── Makefile
│   │   │   ├── memcpy.cu
│   │   │   ├── memcpy.so
│   │   │   ├── README.md
│   │   │   └── test-pipelines.txt
│   │   └── pinhole
│   │       ├── Makefile
│   │       ├── pinhole.cu
│   │       ├── pinhole.so
│   │       └── README.md
│   └── cudamux_algorithms
│       └── mixer
│           ├── Makefile
│           ├── mixer.cu
│           ├── mixer.so
│           ├── README.md
│           └── test-pipelines.txt
└── usr
    ├── include
    │   └── gstreamer-1.0
    │       └── sys
    │           └── cuda
    │               ├── filteralgorithm.hpp
    │               ├── gstcudaallocator.h
    │               ├── gstcudabasefilter.h
    │               ├── gstcudabasemiso.h
    │               ├── gstcudabufferpool.h
    │               ├── gstcuda.h
    │               ├── gstcudameta.h
    │               └── muxalgorithm.hpp
    └── lib
        └── aarch64-linux-gnu
            ├── gstreamer-1.0
            │   ├── libgstcudaexamplefilter.so
            │   ├── libgstcuda.so
            │   ├── libgstopencvfilter.so
            │   └── libgstopencvwarp.so
            ├── libgstcuda-1.0.so -> libgstcuda-1.0.so.0.100.0
            ├── libgstcuda-1.0.so.0 -> libgstcuda-1.0.so.0.100.0
            └── libgstcuda-1.0.so.0.100.0

Eval Binary Installation

To use the binary provided by RidgeRun, export the location of the evaluation binary libraries in your environment.

For Jetson platform

export GST_PLUGIN_PATH=${GST_PLUGIN_PATH}:${PATH_TO_EVALUATION_BINARY}/usr/lib/aarch64-linux-gnu/gstreamer-1.0/
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${PATH_TO_EVALUATION_BINARY}/usr/lib/aarch64-linux-gnu/

where PATH_TO_EVALUATION_BINARY is set the location in your file system where you have stored the binary provided by RidgeRun.

Alternatively, you may just copy the binaries into the standard GStreamer plug-in search path as the follow instructions:

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/lib/aarch64-linux-gnu/gstreamer-1.0/{libgstcudaexamplefilter.so,libgstcuda.so} /usr/lib/aarch64-linux-gnu/gstreamer-1.0

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/lib/aarch64-linux-gnu/{libgstcuda-1.0.so,libgstcuda-1.0.so.0,libgstcuda-1.0.so.0.100.0} /usr/lib/aarch64-linux-gnu

sudo mkdir -p /usr/include/gstreamer-1.0/sys/cuda

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/include/gstreamer-1.0/sys/cuda/{filteralgorithm.hpp,muxalgorithm.hpp,gstcudabasefilter.h,gstcuda.h} /usr/include/gstreamer-1.0/sys/cuda

For x86 platform

export GST_PLUGIN_PATH=${GST_PLUGIN_PATH}:${PATH_TO_EVALUATION_BINARY}/usr/lib/x86_64-linux-gnu/gstreamer-1.0/
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${PATH_TO_EVALUATION_BINARY}/usr/lib/x86_64-linux-gnu/

where PATH_TO_EVALUATION_BINARY is set the location in your file system where you have stored the binary provided by RidgeRun.

Alternatively, you may just copy the binaries into the standard GStreamer plug-in search path as the follow instructions:

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/lib/x86_64-linux-gnu/gstreamer-1.0/{libgstcudaexamplefilter.so,libgstcuda.so} /usr/lib/x86_64-linux-gnu/gstreamer-1.0

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/lib/x86_64-linux-gnu/{libgstcuda-1.0.so,libgstcuda-1.0.so.0,libgstcuda-1.0.so.0.100.0} /usr/lib/x86_64-linux-gnu

sudo mkdir -p /usr/include/gstreamer-1.0/sys/cuda

sudo cp ${PATH_TO_EVALUATION_BINARY}/usr/include/gstreamer-1.0/sys/cuda/{filteralgorithm.hpp,muxalgorithm.hpp,gstcudabasefilter.h,gstcuda.h} /usr/include/gstreamer-1.0/sys/cuda

Eval Binary Testing

Finally, test that the plugin is being properly picked up by GStreamer by running:

gst-inspect-1.0 cuda

with expected output:

Plugin Details:
  Name                     cuda
  Description              Allows frames to be processed by the GPU using a custom CUDA library algorithm
  Filename                 /home/ubuntu/eval-gstcuda/usr/lib/<platform>/gstreamer-1.0/libgstcuda.so
  Version                  0.14.1.1
  License                  Proprietary
  Source module            gst-cuda
  Source release date      2022-09-28 20:35 (UTC)
  Binary package           GStreamer CUDA Plug-in
  Origin URL               Unknown package origin

  cudamux: cudamux
  cudafilter: cudafilter

  2 features:
  +-- 2 elements

gst-inspect-1.0 | grep cuda

with expected output:

cudaexamplefilter:  cudaexamplefilter: GstCUDA Example Filter Element
cuda:  cudafilter: cudafilter
cuda:  cudamux: cudamux

You should see the inspect output for the evaluation binary as shown above.

Example Pipelines

Please refer to the Examples page for reference pipelines.

Troubleshooting

The first level of debug to troubleshoot a failing evaluation binary is to inspect GStreamer debug output.

GST_DEBUG=2 gst-launch-1.0

If the output doesn't help you figure out the problem, please contact support@ridgerun.com with the output of the GStreamer debug and any additional information you consider useful.

If the plug-in is not being recognized, please provide the output of the following command:

gst-inspect-1.0 cuda