cuda-api-wrappers
cuda-api-wrappers copied to clipboard
eyalroz
Thin, unified, C++-flavored wrappers for the CUDA APIs
There are more APIs for working with textures, CUDA arrays and surfaces, than we support. While our array copying support has improved in recent years, our basic representation capability is...
Hi! Thank you for your great work. Do you have a call method with this function? It seems to be used in a different way in the warehouse. That's all...
Let's add comparison operators for classes which are missing them, like contexts - and quite a few others, it seems.
Let's add the matrixMulCUBLAS example program.
With CUDA 12.4, NVIDIA introduced [a new library for writing fatbin files](https://docs.nvidia.com/cuda/nvfatbin/index.html). We need to figure out how to support this (functions on other classes, separate-mostly library etc.) and then...
`cuda::poor_mans_option::value_or` should be a `const` method, and it isn't.
The different vectorAdd examples have different random value generators; and some of them generate `double`s... let's unify all that (and use the standard library, even though it makes the programs...
Our launch config methods for setting block size or grid size in blocks, which take a single value - take the "correct" type: `grid::dimension_t` or `grid::block_dimension_t`. That's a problem, since...
CUDA NVRTC programs have the API function: ``` nvrtcResult nvrtcGetLoweredName( nvrtcProgram prog, const char *const name_expression, const char** lowered_name); ``` which mangle a function's name (or its signature). Unfortunately: *...
unique_span deletion - destruction and de-allocation - is more complex than one might think. With trivially-constructible types, we can supposed just use `new T[size]` to start off and then `delete[]`...