CUDA C++ Libraries Mentorship

I am rapidly up-skilling in CUDA C++ and want an experienced mentor who can walk me through the real-world use of the core foundational libraries—Thrust, CUB, and libcudacxx. My main need is to see clean, well-explained example implementations and concrete use cases rather than abstract theory.

Here is what I have in mind:

• Short, focused code samples that highlight best-practice patterns in each library (device vectors, reductions, custom kernels, cooperative groups, etc.).
• Step-by-step explanations of how these examples map to GPU execution, memory hierarchies, and performance considerations.
• Guidance on how to slot each snippet into an existing CMake-based project so I can experiment immediately.

I already have a CUDA 12.x toolchain set up with Visual Studio and can run tests on an RTX-series GPU. You don’t need to rewrite my code; instead, help me understand the idiomatic way to structure algorithms, manage resources, and chain these libraries together effectively.

The ideal engagement is a mixture of annotated source files plus screen-share sessions where we compile, profile, and tweak together. If you have prior contributions to Thrust, CUB, or libcudacxx—or at least production experience with them—please mention it along with a sample repo or gist I can review.

Let’s start with a small module that covers a typical workflow (e.g., data transfer ➝ transform ➝ reduce) and build outward from there. I’m eager to begin right away and will release milestones once each example compiles, runs, and I fully grasp the reasoning behind it.