i have the following problem: i’m writting a java wrapper for the cuda driver api ( straight forward ) and am planning to load cubins and execute them via this api from within java. the hard part is getting rid on the dependence on the host compiler. i only wish to compile device code but i can’t seem to get this working no matter wheter i use nvcc ( always needs cl on windows no matter the parameters ) or nvopencc aka open64.
did anyone succeed in compiling device code only without the need for a host compiler? i want to get rid of this dependence so i can easily deploy a devkit to my coworkers that does not rely on a c compiler.
edit: whoops, wrong forum. i’m sorry