davideberly/GeometricTools

About

The Geometric Tools Engine (GTE) is a collection of source code for computing in the fields of mathematics, geometry, graphics, image analysis and physics. The engine is written in C++ 14 and supports high-performance computing using CPU multithreading and general purpose GPU programming (GPGPU). Portions of the code are described in various books as well as in PDF documents available at the Geometric Tools Website. GTE is licensed under the Boost Software License 1.0. The update history for the current version of GTE is Current Update History. The update history for all versions of GTE is Full Update History.

I am in the process of writing The Geometric Tools Library (GTL), which is a major revision of GTE. A large portion of GTL development has been porting code from GTE, dealing with size_t versus signed integer compiler complaints, and adding new features. An equally large portion has been writing unit tests and end-to-end tests for the mathematics code. The project has taking a significant amount of time and effort, but the completion is drawing near. GTL is also licensed under the Boost Software License 1.0.

Supported Platforms

The mathematics code is in a header-only library, GTMathematics. A mathematics library with GPU-based implementations is provided, GTMathematicsGPU. The CPU-based common graphics engine code is in its own library, GTGraphics. DirectX 11 wrappers are provided for graphics and applications, GTGraphicsDX11 and GTApplicationsDX11, on Microsoft Windows 10/11. OpenGL 4.5 wrappers are provided for graphics and applications, GTGraphicsGL45 and GTApplicationsGL45, on Microsoft Windows 10/11 and on Linux. A small number of files exist to use GLX and X-Windows on Linux.

On Microsoft Windows 10/11, the code is maintained using Microsoft Visual Studio 2019/2022 with Microsoft's compilers, LLVM clang-cl or with Intel C++ Compiler 2023.

On Ubuntu 22.04.1 LTS, the code is maintained using Visual Studio Code 1.49.2 and CMake 3.15.2, NVIDIA graphics drivers, OpenGL 4.5 and gcc 11.3.0,

On Fedora 38, the code is maintained using Visual Studio Code 1.49.2 and CMake 3.22.2, NVIDIA graphics drivers, OpenGL 4.5 and gcc 13.1.1.

Getting Started

The repository contains many sample applications to illustrate some features of the engine. Top-level solutions/makefiles exist to build everything in the repository. Please read the Installation and Release Notes to understand what is expected of your development environment.

Known 3rd Party Problems

I have had several known problems with compilers I use for testing.

• The "Intel C++ Compiler Classic" is version 19.2 of the compiler. This is now marked as deprecated, so if you select this for the platform toolset to use in Microsoft Visual Studio, you will see deprecation warnings. I had problems with the integration of Intel C++ Compiler 2022 into Microsoft Visual Studio 2022. You enable this compiler by selecting "Intel DPC++/C++ Compiler 2022" as the platform toolset. Warnings occur about macro redefinition, and because I have "treat warnings as errors", the code does not compile. The problem occurred using Intel oneAPI version 2022.3. This has been fixed in Intel oneAPI version 2023.0 where the platform toolset is now "Intel C++ Compiler 2023". It is possible to enable the DPC++ compiler, but not through the MSVS dialogs. I have a tool GeometricTools/Tools/ChangePlatformToolset that will do a batch replacement of the platform toolset in the vcxproj files. The command-line parameter for the DPC++ platform toolset is "Intel(R) oneAPI DPC++ Compiler 2023". However, this compiler does not support __stdcall, so on a Windows machine you can compile only non-Windows-API code (no OpenGL/WGL and no Windows SDK). You can compile the mathematics code you need in your application by generating a library using DPC++ and then linking/loading this into an application that has been compiled with the Intel C++ Compiler 2023.

• The Microsoft Visual Studio 2022 Debugger was broken in versions 17.4.x where x < 5. See the Developer Community explanation for details. The work-around is mentioned there, adding /Zc:nrvo- to the option Configuration | C++ | Command Line. The problem has been fixed as of version 17.4.5.

• After installing NVIDIA driver version 531.18 on a Windows 11 machine with Microsoft Visual Studio 17.5.1, all DirectX 11 samples throw 2 exceptions in D3D11CreateDevice, both tagged as Poco::NotFoundException. If you continue execution after these exceptions, the applications perform correctly. If you then click the application Close icon, another Poco::NotFoundException is generated. You can continue execution after this exception. I have filed a report with NVIDIA, providing a minimal-code reproduction. I recently posted the workaround (listed below) and asked for status on the exception. No response in that forum thread, so I suspect this is low priority for NVIDIA.

  NOTE: Versions 531.29, 531.41, 531.61, 531.68, 531.79, 532.03, and 535.98 of the driver were released and have the same problem.

  WORKAROUND: This assumes you have checked the box in MSVS Exception Settings window that says "<All C++ Exceptions not in this list>". If you do not have this checked, set a breakpoint at a line of code before the call to D3D11CreateDevice and run to the breakpoint. Then check the aforementioned box. Continue the execution and an exception dialog is launched when the call is made to D3D11CreateDevice. That dialog has a box checked "Break when this exception type is thrown". Uncheck the box. You will see a new item in the Exception Settings window under the "C++ Exceptions" that says "Poco::NotFoundException". It is unchecked, which means the debugger will not stop execution for that particular exception. NOTE: This information is stored in the *.user files. If you delete such a file, you lose the exception settings, which means going through the aforementioned process again.

Links to GTE-Based Projects

• Seb Wouter's improvement for my LCP-based test-intersection query between a box and a finite cylinder. When using floating-point arithmetic, the LCP solver had some significant rounding errors to produce incorrect results. His approach clips the box by the cylinder enddisk planes to form a convex polyhedron, projects that to a plane perpendicular to the cylinder axis to obtain a polygon, and then tests for polygon-circle intersection. The code uses a generic convex hull finder to create the convex polyhedron. My current GTE code avoids the generic hull finder and takes advantage of the pairs of parallel planes for the box to amortize the computational costs for a faster algorithm. The repository link is https://github.com/SebWouters/AabbCyl

• CodeReclaimers example of incorporating Dear ImGui into a GTE sample application (for Microsoft Windows). The repository link is https://github.com/CodeReclaimers/GTEImGuiExample