A Bird’s-Eye View of Upcoming System-Level Quantum-Classical Compilers - Unitary Fund

[natestemen]

https://unitary.fund/posts/quantum_compilers.html

[natestemen]

Summary

• "heterogeneous quantum-classical compute node" - really like that phrasing
• compilation tool-chains need solid "intermediate representation" that languages can be lowered to and that can be lowered to native code for many machines/backends
  • IR = description of what you want to do (on a QC) independent of the structure of the source code and independent of how the QC will actually execute the instructions
  • PIRQ has started to do this in late 2021
  • QIR alliance - in linux foundation
• seem to want to build into existing LLVM/compiler tool-chains because they are old (battle tested) and robust and can handle lots of complexity/optimizations
  • not only because they are good though, because of the aforementioned likelihood of "heterogeneous quantum-classical compute nodes". Hence we can get quantum optimizations and classical ones all in one go
• multi-level IR = MLIR. Multiple layers of IR's so that you can do optimizations at each level? I don't really understand this.
• Still in the early days, but because tighter QPU-CPU integration, integrating their optimizations makes more and more sense.

Questions

• What is LLVM?
• OpenQASM as a IR? Or is it something different?
• Maybe checkout qcor (https://qcor.ornl.gov/)

[natestemen]

https://devblogs.microsoft.com/qsharp/introducing-quantum-intermediate-representation-qir/

notes

• QIR specifies a set of rules for representing quantum constructs in LLVM (what is a quantum construct)
• any programming language for gate-based QC can be (presumably) compiled into QIR
• hardware agnostic (so it's truly and intermediate rep)
• it's become more and more clear that near term (and likely down the line) quantum computers will be used in tandem with classical computation. hence the integration of QC tasks in to LLVM
• many source languages can compile into a single IR and optimized and transformed. Once a hardware is known, further compilation takes place to get an executable
• have opened https://github.com/natestemen/mmath/issues/16 to make an example i can use in thesis
• can use clang to compile QIR into executable machine code. If we implement functions in C/C++ to simulate the quantum functions we are using this in effect builds a quantum simulator

[natestemen]

might be worth look at this youtube video to learn about qcor https://www.youtube.com/watch?v=GzslhEnHUHI