I am trying to load the kernels from a cubin file but I want to use only Runtime API and I really can’t find a way to load the kernels with cudaLaunch.
Have anyone managed to solve this problem and write a JCuda program using only the Runtime API?? Is there any other way to do it without using cubin file for the kernels??
Thanks a lot!!
It is in general not possible to launch own kernels with the JCuda runtime API. The JCuda runtime API was mainly intended for the interoperation with JCublas, JCufft and JCudpp. To launch own kernels which are contained in CUBIN files, you have to use the driver API. For that, you may want to use the “KernelLauncher” sample, which simplifies function setup when using the Driver API.