Intermediate representation method and apparatus for compiling computation graphs

An intermediate representation method for compiling computation graphs, comprising the following steps:

• step 1:
• compiling a neural network into a computation graph for neural network computation on a computer;
• step 2:
• constructing a node for each tensor variable in the computation graph;
• step 3:
• associating the node representing the tensor variable in the computation graph to a set of pointers to the tensor variable;
• step 4:
• analyzing constraint relationships between the tensor variables in the computation graph, comprising the following substeps:
• step 4.1:
• defining a constraint representation of an address assignment operation between the tensor variables in the computation graph;
• step 4.2:
• defining a constraint representation of an assignment operation between the tensor variables in the computation graph;
• step 4.3:
• defining a constraint representation of a load operation of the set of pointers to the tensor variable in the computation graph;
• and
• step 4.4:
• defining a constraint representation of a store operation of the set of pointers to the tensor variable in the computation graph;
• step 5:
• iteratively constructing a topological graph of the intermediate representation based on the constraint relationships between the tensor variables in the computation graph, comprising the following substeps:
• step 5.1:
• constructing a propagation process of the intermediate representation of the computation graph based on the constraint representation of the assignment operation between the tensor variables;
• step 5.2:
• constructing a loading process of the set of pointers to the tensor variable in the intermediate representation of the computation graph based on the constraint representation of the load operation of the set of pointers to the tensor variable;
• and
• step 5.3:
• constructing a storage process of the set of pointers to the tensor variable in the intermediate representation of the computation graph based on the constraint representation of the store operation of the set of pointers to the tensor variable;
• and
• step 6:
• analyzing the tensor variables with different aliases pointing to a same memory location based on the intermediate representation, and allocating a register for the tensor variables with different aliases.