Gate Fusion Is Map Fusion
summary
Claim: gate operations are exactly index-transformed maps on the global state vector, so map-map fusion implements gate fusion. Implementation: a Futhark library that expresses gates via flatten/transpose/unflatten vectorization and uses Futhark uniqueness types to update state vectors in place. Correctness follows from tensor/vectorization identities; the Futhark fusion engine merges consecutive gates on the same qubits. Evaluation compares against qsim and QuEST; GPU runs can be up to 10× slower but correctness and fusion are demonstrated.
AI-written summary · about
chapters (8)
- 0:00Introduction to gate fusion
- 1:48Quantum circuit basics
- 6:23Qubit and gate fundamentals
- 12:12Semantics of quantum circuits
- 16:52Imperative simulators vs matrix approach
- 17:13Futhark data parallel language
- 24:09Vectorization and gate fusion
- 28:46Performance and conclusion
related talks
18mLoop Fusion in Matrix Multiplications with Sparse Dependence
55mDeforestation and Program Fusion
17mOptimization and Synthesis of Quantum Circuits with Global Gates
40mSuperfusion: Eliminating Intermediate Data Structures via Inductive Synthesis
45mError Correction with Hybrid-Fusions
19mCompile-Time Java Stream Fusion via mapMulti
16mCompile-Time Java Stream Fusion via mapMulti
32mFault tolerance with the ZX-calculus and fusion complexes: tools for QEC development in the...