Question about kernel implementation Compute() method

[dgiantsidi]

Hi there,

1. What is the difference between contrib_ops and core/providers regarding the kernels' Compute method? What kind of kernels are implemented in the core/providers directories?

2. I want to print all related info about a OpKernelContext object (kernel name, input arguments, output after computation) at the Compute method and all management-related events (cudaMalloc() and other API).

3. Similarly, how can I print all related info about the graph that is constructed after optimizations and right before its scheduling?

Can anyone provide some guidance?

Thanks,
Dimitra