thomasjo / cudalicious

C++ header library intended to reduce CUDA boilerplate code

Cudalicious

C++ header library intended to reduce the amount of CUDA boilerplate code — and hopefully less error-prone.

Huh?

OK, let's assume we have a container with some arbitrary data,

const std::vector<float> foo // ...

If we want to use this data in a CUDA kernel, we can do something like this in order to copy the data from the host to the device

const auto foo_size = sizeof(float) * foo.size();
float* d_foo;
cudaMalloc(&d_foo, foo_size);
cudaMemcpy(d_foo, foo.data(), foo_size, cudaMemcpyHostToDevice);

That's lot of boilerplate code simply to copy some data to the GPU. The observant reader might also have noticed this code does not perform any error checking on those two CUDA functions...

However, if we use Cudalicious, the code above can "magically" be written as

auto d_foo = cuda::copy_to_device(foo);

This is equivalent to the original code, with the added benefit that it also checks for errors ⚡

Usage

Simply grab the header files from the include directory, include the ones you need, and you should be good to go.

See the accompanying examples to get a sense of what your code can look like if you end up giving this library a chance.

Disclaimer

This project is in an early alpha phase and is probably not ready for real production use.

Any and all help is very much appreciated.