cuda-api-wrappers

First - nice writeup which goes into a lot of nooks and crannies. That said, a lot of the user-space "voodoo" is gone if you don't go through CUDA's "runtime API". If you use the driver API, take your kernel source as a string and compile it with NVIDIA's run-time compiler, you'll have better visibility into a lot (not all) of what's going on. For the "raw" version of this, look at: https://github.com/NVIDIA/cuda-samples/tree/master/cpp/0_Int... but for a much more readable, and still fully transparent modern-C++ API version of the same, try this: https://github.com/eyalroz/cuda-api-wrappers/blob/master/exa... that's a sample program for my CUDA API wrappers (header-only) library.

In my library [1], wrapping the CUDA APIs in modern C++, I do allocations which are not exactly from an arena, but something in that neighborhood - memory spaces on context on GPU devices. Unlike the GP suggests, and like you suggest, I have indeed embraced RAII in the library - generally, not just w.r.t. memory allocation. I have not, however, replicated that idioms of the standard library. So, for example: * My allocations are never typed. * The allocation 'primitives' return a memory_region type - essentially a pointer and a size; I discourage the user from manipulating raw pointers. * Instead of unique_ptr's, I encourage the use of unique_span's: owning, typed, lightweight-ish containers - like a fusion of std::span<T> and std::unique_ptr<T[]> . I wonder if that might seem less annoying to GP. --- [1] : https://github.com/eyalroz/cuda