I am reading the paper: Learning to Optimize Tensor Programs.
In the chapter 3.1 Statistical Cost Model, it says:" We extract domain-specific features from a given low-level abstract syntax tree (AST) x. The features include loop structure information (e.g., memory access count and data reuse ratio) and generic annotations (e.g., vectorization, unrolling, thread binding). "
Now, I want to see the features. How can I do?
Sorry for the delayed response; I am currently juggling a few different projects :).
You can probe the features as they are used by looking at the interface used by the TVM XGBoost cost model: e.g., https://github.com/dmlc/tvm/blob/026162ad62ebdfddeb50d8d45278b3c9c781f01c/python/tvm/autotvm/tuner/xgboost_cost_model.py#L328.
1 Like