Efficient linker design method based on multi-thread collaborative optimization

Abstract

The invention relates to an efficient linker design method based on multi-thread collaborative optimization. Firstly, dynamic task division and priority queue scheduling are adopted, the link process is decomposed into a plurality of task units capable of being executed in parallel, and the computing power of a multi-core processor is fully utilized. Secondly, a hierarchical symbol table is introduced, hot symbols are stored in a lockless hash table, cold data are stored in a read-write lock protected red-black tree, thread competition is effectively reduced, and efficient concurrent access is achieved; and meanwhile, a sectional relocation table and a progress counter supported by atomic operation are adopted, so that the concurrency performance is further improved. And finally, generating an intermediate file with a dependency label through a compiler, pre-obtaining a symbol dependency relationship and a dependency graph between modules, avoiding repeated analysis, and designing a dynamic feedback mechanism, so that the linker can feed back optimization data to the compiler to form a closed-loop optimization system. The method is suitable for large-scale projects and high-performance computing scenes, and the resource utilization rate is optimized while the link speed is increased.