How can I get the computation graph of each generated kernel in inductor?
For example, in this code, I have a computation graph and several triton kernels. I want to figure out which triton kernel corresponds to which part of the subgraph.
Or, is it even possible?
My mental model is that inductor breaks the whole computation graph into multiple ones, where each of them can be fused into a single kernel. The process is called scheduling. Please correct me if I’m wrong.
To further clarify: I want to know the part of computation that each fused kernel represents. Otherwise, the wrapper code of fused kernel does not make any sense to me, because I don’t know what it is computing.