I used a temporary tensor to store data in my customized gpu-based op. For debug purpose, I want to print the data of this tensor by traditional printf inside C++. How can I pull this gpu-based tensor to cpu and then print its contents. Thank you very much.
Add a comment |
1 Answer
If by temporary you mean allocate_temp instead of allocate_output, there is no way of fetching the data on the python side.
I usually return the tensor itself during debugging so that a simple sess.run fetches the result. Otherwise, the only way to display the data is the traditional printf inside C++. Given your tensor is an output of your custom operation a tf.Print eases further debugging.
Example:
Tensor temp_tensor; OP_REQUIRES_OK(ctx, ctx->allocate_temp(DT_FLOAT, some.shape(), &temp_tensor)); float* host_memory = new float[some.NumElements()]; cudaMemcpy(host_memory, temp_tensor.flat<Dtype>().data(), some.NumElements() * sizeof(float), cudaMemcpyDeviceToHost); std::cout << host_memory[0] << std::endl; std::cout << host_memory[1] << std::endl; std::cout << host_memory[2] << std::endl; delete[] host_memory; 2 Comments
Charles Tao
Thanks for your kind help. Yes, I want to output the data by traditional printf inside C++. Since it is a gpu-based tensor, I would like to know how to pull the data from gpu to cpu?
Patwie
cumemcpy or
if(!threadIdx.x) printf like in every cuda implementation.