Rapid: Region-Based Pointer Disambiguation
summary
A tool called Rapid uses a profiler + region-based allocator to place hot allocation sites into disjoint regions of 4GB-aligned segments so dynamic checks test top 32 bits of pointers without any memory access. A graph-coloring pass minimizes regions; a round-robin regions policy handles loop-allocated objects. Rapid reports up to 52.94% speedup on Polybench and 1.88% on CPU SPEC 2017, improves 12/16 SPEC benchmarks, with allocator CPU overhead max 0.57% and geometric mean −0.2%.
AI-written summary · about
chapters (10)
- 0:00Introduction to RAPID
- 1:26Core approach
- 3:55Profiling phase
- 6:01Minimizing regions
- 8:10Handling loops
- 10:17Region-based allocation
- 11:41Comparison with Scout
- 12:46Performance benchmarks
- 15:39Conclusion
- 16:02Q&A
related talks
25mThe Road Not Taken: Exploring Alias Analysis Based Optimizations Missed by the Compiler
32mBit-Stealing Made Legal: Compilation for Custom Memory Representations of Algebr...
19mComparing Rapid Type Analysis with Points-To Analysis in GraalVM Native Image
15mPermchecker: A Toolchain for Debugging Memory Managers with Typestate
18mParallelism in a Region Inference Context
17mRecent Developments in Optimizing Compilers for RISC-V
37mHeap Layout Optimisation for Exploitation
16mAccelerate HPC and AI Applications with RVV Auto-Vectorization