The output variable `cshareableHandle_ptr` is incorrectly set to `NULL` and is passed to CUDA, thus always returning `CUDA_ERROR_INVALID_VALUE`. The following similar APIs are likely not working either since they too have output variable `void*`: - cuMemExportToShareableHandle - cuDeviceGetNvSciSyncAttributes - cudaDeviceGetNvSciSyncAttributes - cuMemPoolExportToShareableHandle - cudaMemPoolExportToShareableHandle - cudaGraphMemFreeNodeGetParams - cuMemGetHandleForAddressRange