Automatic and efficient heap data management for limited local memory multicore architectures

Limited Local Memory (LLM) multi-core architectures substitute cache with scratch pad memories (SPM), and therefore have much lower power consumption. As they lack of automatic memory management, programming on such architectures becomes challenging, in the sense that it requires the programmer/compiler to efficiently manage the limited local memory. Managing heap data of the tasks executing in the cores of an LLM multi-core is an important problem. This paper presents a fully automated and efficient scheme for heap data management. Specifically, we propose i) code transformation for automation of heap management, with seamless support for multi-level pointers, and ii) improved data structures to more efficiently manage unlimited heap data. Experimental results on several benchmarks from MiBench demonstrate an average 43% performance improvement over previous approach [1].