Commit af9434b4 authored by Cianciosa, Mark's avatar Cianciosa, Mark
Browse files

Keep the kernel program in memory until the module is built.

parent d37c2bad

graph_framework/cuda_context.hpp

+3 −3
Original line number Diff line number Diff line
@@ -191,12 +191,12 @@ namespace gpu { 
            check_nvrtc_error(nvrtcGetPTXSize(kernel_program, &ptx_size),

                              "nvrtcGetPTXSize");



            check_nvrtc_error(nvrtcDestroyProgram(&kernel_program),

                              "nvrtcDestroyProgram");



            char *ptx = static_cast<char *> (malloc(ptx_size));

            check_nvrtc_error(nvrtcGetPTX(kernel_program, ptx), "nvrtcGetPTX");



            check_nvrtc_error(nvrtcDestroyProgram(&kernel_program),

                              "nvrtcDestroyProgram");



            check_error(cuModuleLoadDataEx(&module, ptx, 0, NULL, NULL), "cuModuleLoadDataEx");



            free(ptx);