Dataflow-Based Optimization for QIR Programs
summary
QDFO preprocesses Microsoft QIR so LLVM’s optimizer can see and apply classical optimizations, then runs a dataflow analysis that eliminates duplicate qubit loads and redundant array create instructions. A preliminary implementation on real QIR shows instruction counts fall as the number of quantum gates grows and LLVM removes more after preprocessing. On a 5‑qubit benchmark QDFO discovers 1 inline and 9 loop/row optimization opportunities and removes repeated pointer loads.
AI-written summary · about
chapters (14)
- 0:00QIR introduction and memory semantics
- 1:33QIR memory IR problems
- 5:42Optimization potential in QIR
- 6:51Two QIR optimization categories
- 7:55Loop unrolling in QIR
- 9:00Redundant loads in QIR
- 10:28Redundant array creation in QIR
- 11:32Preprocessing and optimization workflow
- 12:14Optimization effectiveness validation
- 13:40Results on quantum algorithms
- 14:00Conclusion and future work
- 14:42Q&A: IBM curve flattening
- 15:48Q&A: QIR transformations scope
- 16:51Closing remarks
related talks
29mA Verified Optimizer for Quantum Circuits
20mCompilation of Qubit Circuits to Optimized Qutrit Circuits
23mMake Quantum Computing Useful with Performance Enhancing Infrastructure Software
15mAssertion-Based Optimization of Quantum Programs
40mQ# - going beyond quantum circuits
19mAn Industry Perspective on the Path Forward in Quantum Optimization
24mA Verified Optimizer for Quantum Circuits & Verified Translation Between Low-Level Quantum Languages
15mQOS: Quantum Operating System