Iterative Compilation Method Based on Incremental Instance Learning

For the purpose of making the compiler more adaptive and dealing with complex architecture,a compiler framework is proposed which combines iterative compilation and instance-based learning.On one hand,the compiler can search the optimization space and save the best compiler options into the knowledge library;on the other hand,the compiler can learn from the library to get the best compiler options for the current program.The incremental algorithm can make full use of the accumulated data of the compilation.The algorithms are proposed which can keep the redundant instance out of the knowledge library and filter the noise from the library.