Unleashing the Power of Learning: An Enhanced Learning-Based Approach for Dynamic Binary Translation
Talkat USENIX ATC '19 · on Wednesday, 2019-07-10· at 11:40· USENIX ATC Track II: Grand Ballroom VII–IX·
summary
Enhanced learning extracts translation rules from paired guest/host binaries, uses symbolic execution + SMT-based equivalence checking, and integrates learned rules into QEMU so translation becomes rule matching instead of runtime IR optimization. A sliding-window learning scope (size 2) harvests ~13% more rules, and constrained semantic equivalence (recording unmapped registers as conditions) yields ~23% more; dynamic coverage rises from 55.7% to 69.1% (static 52.2% to 61.8%) and observed speedups reach up to 1.65X with an average of 1.19X.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:24DBT: A Key Enabling Technology
- 1:02Conventional DBT Implementation
- 2:50Recent Advances: Learning-Based Approach
- 4:05Low Translation Coverage
- 5:04Limitation : Limited Learning Scope
- 6:40Limitation II: Strict Equivalence Verification
- 10:01Solution Il: Constrained Semantic Equivalence
- 11:15Experimental setup
- 12:28Performance Speedup
- 15:19Solution : Variable Learning Scope
related talks
17mAccelerating Shared Library Execution in a DBT
59mCompiler-assisted Translation Validation
17mCrossMapping: Harmonizing Memory Consistency in Cross-ISA Binary Translation
15mKeynote: Bridging The Divide: Unifying RISC-V Through Binary Translation
27mImproving Vectorization Heuristics in a Dynamic Compiler with Machine Learning Models
54mProgramming Language Design and Implementation for the Machine Learning Era: A Personal(…)
15mJavART: a Lightweight Rule-Based JIT Compiler Using Translation Rules Extracted from(…)
11mQiMeng-Xpiler: Transcompiling Tensor Programs for Deep Learning Systems with a...