dyanab
1
Now I have a cubin file, target at Turing (sm_75).
I loaded it with runtime api:
cuModuleLoad(&module, "a.cubin");
cuModuleGetFunction(&kernel, module, "kenrel");
The problem is I cannot set function attributes with cudaFuncSetAttributes. It returns 98 (cudaErrorInvalidDeviceFunction).
How can I fix this?
Thanks!
dyanab
2
BTW, I can launch this kernel with cuLaunchKernel and get the correct result.
There is no function cudaFuncSetAttributes
I assume you meant cudaFuncSetAttribute
It’s evident you’re using the driver API. Why would you switch to the runtime API for this?
Did you try cuFuncSetAttribute?
https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__EXEC.html#group__CUDA__EXEC
dyanab
4
Hi @Robert_Crovella,
Switching to cuFuncSetAttribute solved this issue.
I just came across an example using cudaFuncSetAttribute, and did not aware of the difference between runtime API and driver API.
Thank you.