[RuntimeError] CUDA error: no kernel image is available for execution on the device

[Pooookeman]

Thank you for your brilliant CUDA kernel of sparse spiking BP. But I got a RunTimeError when executing the CUDA code.
The error happened in line 459 "train_and_time.py"
grad_weights = s3gd_backward_cuda(spk_trace, aout_b, aout_t, aout_i, ds_out, nb_inputs, nb_hidden)
While the CUDA kernel is fine when running other commands like print(torch.tensor([1.0, 2.0]).cuda()).
Then get tensor([1., 2.], device='cuda:0') as the result.
And my pytorch (= 1.7.1) and cuda (=11.0) are identical to your setting.
Do you have any idea about the reason for this problem?? Since it only happens in the s3gd kernel, so I thought that might be some problem hidden in the CUDA code? Could you provide some hints for this error?