HIP RT SDK

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, 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

Requirements

HIP RT runs on AMD and NVIDIA GPUs. HIP and CUDA© APIs are dynamically loaded so you do not need to have these SDKs if your have these DLLs installed by the GPU driver package. Hardware accelerated ray tracing only works on AMD RDNA2-based GPUs.

It works on any AMD GPU which supports HIP. The supported AMD GPU architecture families are:

  • Navi2x (Radeon™ RX 6000 series)

  • Navi1x (Radeon™ RX 5000 series)

  • Vega2x

  • Vega1x

You need an AMDGPU driver newer than 21.40. However, we recommend using 21.50 or newer.

Directories

  • hiprt

    The header and the library which you need to download from HIP RT product page

  • tutorials

    Tutorial source code

  • contrib

    External dependencies

Building the Tutorial

  1. Сlone the repository, then init, and update the submodules if you didn’t clone with --recursive.

git submodule init
git submodule update

  1. Download the HIP RT SDK from HIP RT product page, copy hiprt directory to here.

  2. Run premake like this on Windows, which will generate a solution for Visual Studio 2019:

cd tutorials
../tools/premake5/win/premake5.exe vs2019

  1. Open the solution, compile & run. These tutorials are made to run on both AMD and NVIDIA by specifying the device index.

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

In This Section