Overview of the autograph framework. (IMAGE)
Caption
Autograph first extracts loops and builds dependency graphs capturing instruction semantics and data flow, which are then converted into embeddings by Graph Neural Network. These embeddings are then fed to a deep reinforcement learning agent that predicts the best vectorization and interleaving factors, injects the corresponding pragmas, runs the code, and uses the runtime as a reward to improve its predictions.
Credit
Yao Xiao et al.
Usage Restrictions
Credit must be given to the creator.
License
CC BY