This repos demonstrates OpenCL GPGPU and its extensive programing skills, which include but not limited to following objectives:

• Algebra solutions, such as matrix transpose(in-place or out-place), matrix multiplication, matrix-vector multiplication, tridiagonal system solvers(Thomas, CR, PCR, RD and their hybrids);
• grayscale image histogram solution;
• OpenCL-OpenGL interoperation examples: simulating heat transfer equation and visualizing the results, Depth of View Blur sample(Solving discreted heat diffusion equation and visualizing the results);
• More examples will be coming soon.

• VC++ compiler support C++17 or higher, install Visual studio 2017 and above with Visual C++ workload is recommanded;
• cmake with version 3.19.* or above;
• OpenCL SDK with OpenCL Specification 2.0 or above installed:
  Intel OpenCL SDK is a common choise for desktop NVIDIA/AMD GPU, other SDKs from GPU venders may also be applicable(I have not tested them yet), Khronos Group OpenCL SDK is also a alternative choise(Already tested).
• Third party repos must be placed in the parent directory of this repos's local copy with subdirectory name "ThirdParty". Third party reposes must provide OpenGL development utlities:
  • glad, OpenCL Version 4.6 or above, Core Profile.
  • glm
  • glfw
  • spirv-1.1
• Optional development tools: Visual Studio Code with extensions: C/C++, CMake, CMake Tools, OpenCL, Output Colorizer(These is my preferred development evironment). Alternatively, you can use Visual Studio on windows if CMake support workload has been included in your installation.

Following standard cmake build procedure if you use VSCode(Check Set CMAKE_BUILD_TYPE also on multi config generators in Settings), or you need the following additionall settings(Again, I suggest you to use VSCode):

• Before build debug version, define CMAKE_BUILD_TYPE=Debug in your cmake cache(CMakeCache.txt), and just kick cmake default build procedure;
• Before build release version, select cmake variant to Release, then edit CMakeCache.txt with the option:CMAKE_BUILD_TYPE=Release, then kick off cmake build procedure.

I can only guarantee all the examples will be work properly for AMD GPU, as for NVIDIA GPU, the following known issues must be solved if you are trapped by a debug asset or objectionable wrong outputs:

• Fill arguments corresponding to local_work_size in API calls clEnqueueNDRangeKernel;
• In file matrix/mat_mul_vec.cl, synchronize all the local tile reduction writes operations;