A TENSOR PRODUCT COMPILER FOR SCALABLE PARALLEL COMPUTERS

THE OBJECTIVE OF THIS WORK IS TO DEVELOP A TENSOR PRODUCT COMPILER THAT CAN PARSE TENSOR PRODUCT FORMULATIONS INTO WORKING CODE FOR SCALABLE PARALLEL PROCESSORS. THE TENSOR PRODUCT COMPILER WILL USE MATHEMATICAL TRANSFORM RULES TO OPTIMIZE CODE FOR SUCH PARAMETERS AS NUMBER OF PROCESSORS, THE ARCHITECTURE OF THE NODE PROCESSORS, THE RELATIONSHIP BETWEEN LOCAL AND GLOBAL MEMORY, AND LATENCY AND BANDWIDTH OF INTERPROCESSOR COMMUNICATION. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - THE TENSOR PRODUCT COMPILER WILL USE MATHEMATICAL TRANSFORM RULES WHEN OPTIMIZING CODE, THUS MAKING THESE COMPUTERS MUCH EASIER TO PROGRAM. AS THESE DIFFICULTIES ARE RESOLVED, SCALABLE MULTI-PROCESSORS SUCH AS THESE WILL BECOME VIABLE FOR MAINSTREAM BUSINESS COMPUTING.