Systems and methods for automatic computer code parallelization

A system that automates computer code parallelization, the system comprising:

• optimizable code detection circuit that:
• measures a performance of a computer program includes identifying performance metrics for one or more disparate blocks of code of the computer program;
• identifies a subset of the one or more disparate blocks of code that do not satisfy a performance threshold;
• dependence analysis circuit that performs a dependence analysis of the subset and collects dependence data resulting from the dependence analysis;
• side effect analysis circuit that performs a side effect analysis of the subset and collects side effect data resulting from the side effect analysis;
• one or more computer processing circuits that:
• construct a dependency analysis basic block (DABB) graph for each disparate block of code of the subset of the one or more disparate blocks of code, wherein the DABB graph for each disparate block code of the subset comprises:
• a graphical representation of one or more possible paths through a respective disparate block of code of the subset of the one or more disparate blocks of code, wherein the graphical representation of the one or more possible paths is generated based on a list of code instructions derived from a control flow graph of the computer program that corresponds to the respective disparate block of code of the subset;
• one or more annotations indicating an ability to thread the respective disparate block of code of the subset based on the dependence data and the side effect data;
• construct a versioned dependency graph that optimizes a performance of the computer program, wherein constructing the versioned dependency graph includes:
• integrating the DABB graph for each disparate block of code of the subset within the control flow graph of the computer program;
• injecting a metaprogram into the computer program, wherein the metaprogram is based on the versioned dependency graph;
• and
• automatically executes parallelization of the computer program at runtime based on the metaprogram.

A computer-implemented method for constructing a versioned dependency graph that enables automated computer code parallelization, the computer-implemented method comprising:

• constructing a versioned dependency graph that optimizes a performance of a computer program, wherein constructing the versioned dependency graph includes:
• (i) implementing a computer code analysis of one or more disparate blocks of code of the computer program, wherein implementing the computer code analysis includes:
• implementing a dependence analysis of the one or more disparate blocks of code and collecting dependence data resulting from the dependence analysis;
• implementing a side effect analysis of the one or more disparate blocks of code and collecting side effect data resulting from the side effect analysis;
• (ii) constructing a dependency analysis basic block (DABB) graph for each of the one or more disparate blocks of code of the computer program, wherein the DABB graph for each of the one or more disparate blocks code comprise:
• a graphical representation of one or more possible paths through a respective disparate block of code of the one or more disparate blocks of code, wherein the graphical representation of the one or more possible paths is generated based on a list of code instructions derived from a control flow graph of the computer program that corresponds to the respective disparate block of code;
• one or more annotations indicating an ability to thread the respective disparate block of code based on the dependence data and the side effect data;
• (iii) integrating the DABB graph for each of the one or more disparate blocks of code within the control flow graph of the computer program;
• (iv) injecting a metaprogram into the computer program, wherein the metaprogram is based on the versioned dependency graph;
• and
• (iv) automatically executing parallelization of the computer program at runtime based on the metaprogram.