Hi,
I have worked CUDA program, and I want to compile CUDA part to binary for all possible architectures sm_10, sm_11, sm_13 and have the binaries in my C sources. So I can give an algorithm in sources (Linux/Windows, 32/64 bit independent) but all my CUDA algorithms are already compiled and does not exist in sources.
I generated several fatbin files (main.fatbin.10.c, main.fatbin.11.c, main.fatbin.13.c) from main.cu. Now I need to bind the kernels to my main program. If I just take the source and link it together with main.fatbin.XX.c I got unresolved kernels.
Please, sugget me what I did wrong.
Thank you!
Ilghiz