Rapid Quantum Compiler Prototyping through a Multi-Level Intermediate Representation

As quantum-classical computing architectures progress toward tighter CPU-QPU integration, programmers of these novel accelerated systems will need robust and performant compiler infrastructures that map high-level programmatic representations to a heterogeneous set of binary code and executables. Moreover, there is a desire to see tight integration with classical compiler systems in an effort to promote future coupling of classical and quantum code bases. Recently, researchers have defined a quantum-classical runtime API specification based on the LLVM intermediate representation (IR). This Quantum Intermediate Representation (QIR) provides a unified abstraction layer for language compilers as well as robust code generation utilities for available quantum computer backends and simulators. We present a compiler platform, qcor, that lowers language representations to LLVM code adherent to the QIR specification. Our approach leverages the novel Multi-Level Intermediate Representation (MLIR) framework to take a language-level quantum IR tree down to the QIR. We leverage this multi-level representation for pertinent quantum optimizations and circuit synthesis. Ultimately, our approach enables quick mapping of quantum language parse trees to the MLIR, which can be readily lowered to the QIR for executable generation. We will demonstrate the utility of this approach, including how it enables multi-quantum-language integration and library development.