CUDA Plugin crashes in release mode on cuMemcpyDtoDAsync

Hi everyone, I have been working on a CUDA plugin for a while now and I have to make a release build to send it to my employer for testing, in the beggining I couldn't get it to even compile but it does compile now thankfully. The issue is now I am getting CUDA errors relating mostly to memory on release, I'm not sure why, the project is based off of SDK_Invert_Procamp from the 2022 SDK so the compile flags for the kernel are mostly identical with that except the MSVC version of course for cl.exe. I am getting issues even with things like cuMemcpyDtoDAsync and cuMemFreeHost with path emission.

When I am testing other parts of the plugin, I again get an Illegal Memory Address 700 error after running the update kernel.

Or here, where I call cudaMemset on the screen tiles:

Should I move all the CUDA specific code to the .cu file? That might be the solution after all but I would rather not do that. The code works perfectly with no crashes in debug mode. just to be clear.