An interactive, cross platform volume renderer based on the OpenCL compute API. More specifically, a front-to-back ray casting algorithm with regular step size is used to evaluate an emission/absoption model for voxel data in regular grids (scalar density field). Alternatively, a path tracer based on Woodcock tracking may be used for rendering (experimental). The volume renderer features early ray termination, object order (and image order) empty space skipping, local illumination, and various gradient based shading techniques. The rederer is designed to run interactive on the GPU in single node environments. The data set size is limited by available GPU memory. Execution on CPU is possible but not recommended due to severe performance issues.

The code is structured as followed:

• kernel: OpenCL C parallel volume rendering kernel
• core: C++ interface to volume rendering kernel
• oclutils: utilities for setting up OpenCL and OpenCL-OpenGL interop
• io: volume data file reader (at the moment only dat/raw format)
• qt: everything GUI related: OpenGL screen quad rendering, mouse/keyboard interaction, transfer function editor, color picker, parameter controls, histogram rendering...

This renderer is primarily used as a basis for my research projects. Therefore, the code is partly lacking documentation, automated testing, and portability. However, I'm trying to write clean and understandable code as well as testing on different systems.

• Qt 5.12 or later for the GUI.
• CMake 3.9 or later for building.
• C++17 compiler.
• OpenCL 1.2 (capable device & drivers, headers, C++ bindings)
• OpenGL 3.3 or later for dispaying the texture generated with OpenCL.
• OpenMP 2.0 or later (3.1 or later recommended for faster loading times).

The renderer can display volume (timeseries) data sets. An excellent resource for those is Pavol Klacansky's Open Scientific Visualization Datasets page. For data loading you should create and select a .dat file containing a description of the raw (binary) volume data following this scheme:

ObjectFileName: 	relative/path/to/binaryData.raw
Resolution: 		256 256 256 
SliceThickness:		1.0 1.0 1.0
Format: 		UCHAR

The ObjectFileName may contain multiple paths to different time steps. Alternatively, Resolution may be extended with a fourth dimension if the raw file names of the timesteps contain a suffix with ascending numbering. Currently supported formats are: UCHAR, USHORT, and FLOAT.

Optional qualifyers:

• Endianness: can be set to LITTLE (default) or BIG.
• ChannelOrder: can be set to R (default) for single value scalar fields, RGBA for pre-computed color and opacity fields, or RG for two channel scalar fields (single color and opacity).

Use CMake to build the volume renderer:

git clone https://github.com/vbruder/VolumeRendererCL.git
cd VolumeRendererCL
mkdir build
cd build
cmake .. -DCMAKE_PREFIX_PATH=/path/to/Qt/install
make -j `nproc`

Make sure to replace the CMAKE_PREFIX_PATH with the path to your Qt install directory, e.g. /home/username/Qt/5.11.2/gcc_64/

Currently, I develop and test on an Ubuntu 18.04 based Linux using GCC 7.4.0, the latest stable version of Qt and an NVIDIA Titan X (Pascal). Sporadically, I try to also test on a Windows 10 system using the Visual Studio 17 compiler and an AMD Vega FE, as well as on my integrated Intel Gen9 GPU (NEO driver stack).

Systems/hardware I successfully ran this renderer on:

• NVIDIA Maxwell & Pascal, AMD Fiji & Vega, Intel Gen9 GPU & Skylake CPU
• GCC 7.4.0, Visual Studio 2017 (v140), Clang 6.0
• Qt 5.12.2 & Qt 5.13.0
• CMake 3.10.2 & 3.12.2

Copyright (C) 2017-2019 Valentin Bruder [email protected]

This software is licensed under LGPLv3+.

• Color wheel from Mattia Basaglia's Qt-Color-Widgets: https://github.com/mbasaglia/Qt-Color-Widgets
• OpenCL utils based on Erik Smistad's OpenCLUtilityLibrary: https://github.com/smistad/OpenCLUtilityLibrary
• Transfer function editor based on Qt sample code.