HIP RT SDK

HIP RT tutorials. Premake automatically builds HIPRT from its public source repository when hiprt/hiprt.h is missing. Existing installations are preserved.

HIP RT is a ray tracing library in HIP. The APIs are designed to be minimal and low level, making it easy to write a ray tracing application in HIP. We designed the library and APIs to be simple to use and integrate into any existing HIP applications. Although there are a few other ray tracing APIs which, we designed HIP RT to be simpler and easier to use, so you do not need to learn many new kernel types.

Features

• Ray-triangle intersection
• Ray-custom primitive intersection
• Ray mask to filter geometries
• Several bounding volume hierarchy (BVH) options
• Offline BVH load and store
• BVH import
• Motion blur

Requirement

HIP RT runs on AMD and NVIDIA® GPUs. HIP and CUDA® APIs are loaded dynamically at runtime. Building HIP RT from source with the scripts requires the GPU SDK and build tools listed under Building the tutorials. Hardware-accelerated ray tracing only works on RDNA™ 2 GPUs (Radeon™ RX 6000 series or newer).

The supported AMD GPU architecture families are:

• Navi4x (Radeon™ RX 9000 series)
• Navi3x (Radeon™ RX 7000 series)
• Navi2x (Radeon™ RX 6000 series)
• Navi1x (Radeon™ RX 5000 series)

You will also need AMD Software: Adrenalin Edition 23.30 or later for Windows and ROCm 5.7 for Linux.

---

Directories

• hiprt
  • Headers, library and GPU kernels installed by the source build scripts. See details.
• tutorials
  • Tutorial source code
• contrib
  • External dependencies

Building the tutorials

Install Git, Python 3.8+, CMake 3.19+ and HIP SDK / ROCm with hipcc and HIPRTC. Windows requires Visual Studio 2022 with the C++ tools; Linux requires a C++17 compiler and Make. Set HIP_PATH to your HIP installation (default on Linux: /opt/rocm).

Clone the repository, then run the commands below from its root. Premake initializes missing submodules and builds HIPRT if needed.

Windows

cd tutorials
..\tools\premake5\win\premake5.exe vs2022

Open tutorials/hiprtSdkTutorial.sln in Visual Studio 2022 and build Release / x64. Then, from tutorials, run a demo:

cd dist
bin\Release\19_primary_ray64.exe

Linux

chmod +x tools/premake5/linux64/premake5
cd tutorials
../tools/premake5/linux64/premake5 gmake2
make config=release_x64 -j
cd dist
export LD_LIBRARY_PATH="../../hiprt/linux64:${HIP_PATH:-/opt/rocm}/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}"
bin/Release/19_primary_ray64

Introduction to the HIP RT APIs

The minimum example can be found at tutorials/00_context_creation/main.cpp. On AMD platforms, you need to create a HIP context and device to initialize hiprt using hiprtCreateContext by passing hiprtContextCreationInput object where you need to set the native API context, device, and device type (HIP or CUDA).

After that, use hiprtCreateGeometry and hiprtBuildGeometry to create hiprtGeometry. Once you have finished setting up objects on the host side, you can pass the context object to your kernel.

An example of a minimum kernel can be found at here. This is a simple HIP kernel that we are passing hiprtGeometry to. To perform an intersection test, simply fill in hiprtRay, then create a hiprtGeomTraversalClosest object, then call getNextHit(). It is that simple.

References

• Introducing HIP RT – a ray tracing library in HIP, GPUOpen Blog
• HIP Ray Tracing 2.0 introduces improved custom function handling, Bitcode linking, and more, GPUOpen Blog
• Introducing HIP RT v2.1 - batch construction for small geometries, transformation query functions, and more, GPUOpen Blog
• Introducing HIP RT v2.2, GPUOpen Blog