Compiler transform optimization for non-local functions
Systems and methods for using compiler transforms to transform a non-local function into a local function are disclosed. The systems and methods perform a dynamic inter-procedural analysis before performing reverse-mode automatic differentiation. The dynamic inter-procedural analysis is performed to determine a maximum set of computer program information. A non-local to local transformation is applied to the determined maximum set of computer program information, and each original instruction is mapped to an optic that is represented as an opaque closure in the transformed local function.
1. A method for performing a compiler transform on code that represents a computer program written in an existing programming language, the method comprising:
receiving, by a graphical user interface, the code from a user that represents the computer program written in the existing programming language, wherein the code includes a plurality of original instructions;
optimizing, by a compiler, the compiler transform on the code that represents the computer program by performing a dynamic inter-procedural analysis of the computer program to determine a maximum set of computer program information that can be determined from the computer program during execution of the computer program;
applying, by the compiler, a non-local to local transformation on the determined maximum set of computer program information to generate a transformed local function;
mapping, by the compiler, each of one or more original instructions of the plurality of original instructions to an optic having a composition property that allows the optic to be combined with another optic;
implementing, by the compiler, a representation of the optic as an opaque closure in the transformed local function; and
generating, by the compiler, a transformed computer program that represents the transformed local function, wherein the transformed computer program is executed to generate a derivative of the computer program.
2. The method of claim 1 , wherein the mapping includes treating a primal function as the optic.
3. The method of claim 1 , wherein the code that represents the computer program defines a mathematical model of a function.
4. The method of claim 1 , wherein the dynamic inter-procedural analysis of the computer program is performed through lattice-based data-flow analysis.
5. The method of claim 1 , wherein the computer program is evaluated on an abstract symbolic domain.
6. The method of claim 1 , wherein the method is applied to a physics-informed neural network or a physics-informed generative adversarial network.
7. The method of claim 1 , wherein the non-local to local transformation includes interleaving a transformation step and an optimization step.
8. The method of claim 7 , wherein the non-local to local transformation is delayed until at least one optimization step has been performed, and wherein, after the at least one optimization step has been performed, the transformed computer program is generated for an n th -order transformation.
9. The method of claim 1 , wherein the non-local to local transformation includes creating a data structure for an n th -order residual such that the non-local to local transformation can be optimized.
10. A system for performing a compiler transform on code that represents a computer program written in an existing programming language, the system comprising:
a computer having a processor that executes a compiler and a graphical user interface, wherein the graphical user interface is adapted to receive the code from a user that represents the computer program written in the existing programming language, wherein the code includes a plurality of original instructions, and wherein the processor is adapted to cause the compiler to:
optimize the compiler transform on the code that represents the computer program by performing a dynamic inter-procedural analysis of the computer program to determine a maximum set of computer program information that can be determined from the computer program during execution of the computer program;
apply a non-local to local transformation on the determined maximum set of computer program information to generate a transformed local function;
map each of one or more original instructions of the plurality of original instructions to an optic having a composition property that allows the optic to be combined with another optic;
implement a representation of the optic as an opaque closure in the transformed local function; and
generate a transformed computer program that represents the transformed local function, wherein the transformed computer program is executed to generate a derivative of the computer program.
11. The system of claim 10 , wherein the mapping includes treating a primal function as the optic.
12. The system of claim 10 , wherein the code that represents the computer program defines a mathematical model of a function.
13. The system of claim 10 , wherein the dynamic inter-procedural analysis of the computer program is performed through lattice-based data-flow analysis.
14. The system of claim 10 , wherein the computer program is evaluated on an abstract symbolic domain.
15. The system of claim 10 , wherein the method is applied to a physics-informed neural network or a physics-informed generative adversarial network.
16. The system of claim 10 , wherein the non-local to local transformation includes interleaving a transformation step and an optimization step.
17. The system of claim 16 , wherein the non-local to local transformation is delayed until at least one optimization step has been performed, and wherein, after the at least one optimization step has been performed, the transformed computer program is generated for an n th -order transformation.
18. The system of claim 10 , wherein the non-local to local transformation includes creating a data structure for an n th -order residual such that the non-local to local transformation can be optimized.