Transparent pointer compression for linked data structures

64-bit address spaces are increasingly important for modern applications, but they come at a price: pointers use twice as much memory, reducing the effective cache capacity and memory bandwidth of the system (compared to 32-bit address spaces). This paper presents a sophisticated, automatic transformation that shrinks pointers from 64-bits to 32-bits. The approach is “macroscopic,” i.e., it operates on an entire logical data structure in the program at a time. It allows an individual data structure instance or even a subset thereof to grow up to 232 bytes in size, and can compress pointers to some data structures but not others. Together, these properties allow efficient usage of a large (64-bit) address space. We also describe (but have not implemented) a dynamic version of the technique that can transparently expand the pointers in an individual data structure if it exceeds the 4GB limit. For a collection of pointer-intensive benchmarks, we show that the transformation reduces ...

• Type: Conference
• Year: 2005
• Where: ACMMSP
• Authors: Chris Lattner, Vikram S. Adve
• Keywords: ACMMSP 2005, Address Space, Data Structure, Individual Data Structure