vyom1611/learning-cuda-trt

A large number of cuda/tensorrt cases .

Learning CUDA and TensorRT

A large number of CUDA/TensorRT examples are provided in this project.

• CUDA Driver API
• CUDA Runtime API
• Introduction to TensorRT Basics
• Basic TensorRT Learning
• Plugins and ONNX Parser
• Advanced TensorRT Applications
• Exporting ONNX Models, Preprocessing, and Postprocessing
• Handling Complex Situations with Specific Project Examples

This is a source code project for learning CUDA and TensorRT through examples. It covers a wide range of cases, from basic CUDA driver and runtime APIs to introductory and advanced TensorRT concepts. It also includes model export, preprocessing, postprocessing, and multithreading encapsulation.

Usage Instructions - Manual Environment Configuration

• All examples use Makefile as the build tool.
• Special variables like ${@CUDA_HOME} should be replaced with your system's actual environment values.
• After configuration, you can usually compile and run with make run.

Usage Instructions - Automatic Environment Configuration

• Requires a Linux system (Ubuntu 16.04 or higher) with GPU and NVIDIA driver version 495 or higher.
• Install the Python package: pip install trtpy -U -i https://pypi.org/simple
• Set up shortcut: echo alias trtpy="python -m trtpy" >> ~/.bashrc
• Apply shortcut: source ~/.bashrc
• Configure key: trtpy set-key sxaikiwik
• Fetch and configure environment: trtpy get-env --cuda=11 (supports CUDA versions 10 and 11 only)
• Automatically update configuration variables: trtpy prep-vars . to replace all variables in the current directory.
• Run with make run

References

• TensorRT Bilibili Video Explanation
• Official Video Explanation
• TRTPy Introduction Document
• Video Tutorial for This Source Code (Tencent Classroom)