Session F27: Programming and Compilation -- the Quantum Computing Stack

The quantum computing stack: From quantum algorithms to optimized resource estimates

It is known that quantum computers offer up to exponential speedups over their classical counterparts for solving certain computational tasks. However, concrete comparisons of resource requirements for specific problem instances are still scarce. In this talk, I will discuss how software for quantum computing can support researchers in carrying out such analyses. Specifically, I will focus on automatic compiler optimizations and how these may offer great benefits even in the presence of hand-optimized libraries.   

Automatic Compilation for Portable and Scalable Quantum Software

Noisy intermediate-scale quantum (NISQ) processors are becoming larger (20–150 qubits) and more expressive (e.g., supporting a plurality of two-qubit interactions). However, since NISQ processors are error-prone, there is a great disparity between those programs which are desirable to write and those which execute reliably. Together, these aspects make it difficult to write quantum programs which are both portable (i.e., readily executable on different devices) and efficient (i.e., produce more accurate results using fewer resources). We discuss how automatic compilation helps achieve these simultaneous goals more consistently, and compare results of an open-source, extensible automatic compiler quilc to semi-manual counterparts.   

Compiler tools for hybrid quantum-classical algorithms

We describe the Rigetti compilation toolchain and in particular how it supports optimized implementations of certain hybrid quantum-classical algorithms. Programs written in Quil are transpiled into a restricted subset of Quil instructions that are realizable on the available control hardware and target chip topology. These transpiled programs are further compiled into binary executables for custom FPGA pulse sequencers. The toolchain provides two key features that enable high performance hybrid computing: (1) gate parameters from the original input Quil program are translated to sequencer instructions that load from classical memory shared between the sequencer and classical host computer; (2) compiled programs can contain arbitrary control flow that branches off of single-qubit measurement results. The first feature enables the compilation of Quil into binaries that can be updated at run-time and the second, enables active reset of qubit states. Together these allow for rapid iteration in applications such as the optimization of a variational quantum algorithm, because these binaries can be re-executed many times for different input parameters without need for re-compiling or waiting for qubits to relax. We provide quantitative benchmarks of the improved wall-clock performance.   

Quantum Circuit Compilation to NISQ processors

We describe automated reasoning approaches for QCC-NISQ, quantum circuit compilation to NISQ (noisy intermediate-scale quantum) processor architectures. We tested the approaches for different NISQ processor architectures and QAOA (quantum alternating operator ansatz) circuits. This approach is integrated into our software suite for automated, architecture aware, compilation for emerging gate-model quantum computers. We give an overview of the key components of this suite: a circuit synthesizer, a QCC-NISQ solver, and a visualizer.   

Less than a million CNOTs should be enough to solve a classically intractable instance of a scientific problem with a quantum circuit

The question I will try to answer in this talk is the following: what is the size of the smallest quantum computation capable of solving an instance of a scientifically interesting problem that is intractable for a classical computer? The problem I consider is Hamiltonian dynamics simulation, in the sense of the ability to sample probability distribution given by a state evolved under the target Hamiltonian for a specified time t and accurate to within a specified error epsilon. The core of the talk concerns the development and application of a range of algorithm design, circuit optimization, and hardware/software co-design techniques, the combination of which leads to the reduction in the quantum resource counts provided by pure algorithmic formulations by several orders of magnitude. Specifically, the best physical-level quantum circuit features under 650,000 CNOT gates, and the best fault-tolerant circuit features under 6.8x10^6 T gates. This illustrates that algorithmic and software-level optimizations will be indispensable for practical quantum computing.   

Overview and Comparison of Gate Level Quantum Software Platforms

Quantum computers are available to use over the cloud, but the recent explosion of quantum software platforms can be overwhelming for those deciding on which to use. In this paper, we provide a current picture of the rapidly evolving quantum computing landscape by comparing four software platforms--Forest (pyQuil), QISKit, ProjectQ, and the Quantum Developer Kit--that enable researchers to use real and simulated quantum devices. Our analysis covers requirements and installation, language syntax through example programs, library support, and quantum simulator capabilities for each platform. For platforms that have quantum computer support, we compare hardware, quantum assembly languages, and quantum compilers. We conclude by covering features of each and briefly mentioning other quantum computing software packages.   

A case study for quantum software development: Linear systems solver

As the power of quantum computing hardware is growing, software implementations and tools for accurate resource estimation that goes beyond simple asymptotic scaling become crucial for judging the feasibility of an application. Solving linear systems of equations is one of the prime applications for which quantum algorithms with an exponential advantage over the best known classical algorithms have been developed. In this talk, we report on the implementation of state-of-the-art quantum algorithms solving partial differential equations within the Microsoft Quantum Development Kit, which provides robust simulation, debugging, and resource estimation facilities. A key point in this work has been the handling of arithmetic functions: Gate synthesis facilitates accurate resource counts and is a requisite for deployment to quantum hardware, whereas emulation by the classical simulator allows for software testing on larger systems. To this end, we have equipped the Microsoft Quantum Development Kit with a generic and efficient emulation capability for quantum oracles defined by classical functions.   

Two-step approach to scheduling quantum circuits

As the effort to scale up existing quantum hardware proceeds, it becomes necessary to schedule quantum gates in a way that minimizes the number of operations. There are three constraints that have to be satisfied: the logical dependency of the quantum gates in the algorithm, the fact that any qubit may be involved in at most one gate at a time, and the restriction that two-qubit gates are implementable only between connected qubits. The last aspect implies that the compilation depends not only on the algorithm, but also on hardware properties like connectivity. Here we present a two-step approach in which logical gates are initially scheduled neglecting connectivity considerations and routing operations are added at a later step in ways that minimize their overhead. We rephrase the subtasks of gate scheduling in terms of graph problems like edge-coloring and maximum subgraph isomorphism. While this approach is general, we specialize to a one-dimensional array of qubits to propose a routing scheme that is minimal in the number of exchange operations. As a practical application, we schedule the quantum approximate optimization algorithm in a linear geometry and quantify the reduction in the number of gates and circuit depth that results from different scheduling strategies.   

Real-time randomized compilation of quantum algorithms

Recent experiments have indicated the effectiveness of Pauli Frame Randomization (PFR) as a noise shaping technique. Not only does PFR tailor general noise into Pauli-stochastic noise, it also decouples qubits from their noise environment increasing system Markovianity. Presently, the randomization process requires extensive pre-compilation of pulse sequences. Here, we demonstrate an in hardware implementation of PFR on the BBN Arbitrary Pulse Sequencer II that generates randomized sequences in real-time. We first implement randomized benchmarking without pre-computing pulse sequences before using this capability to randomize a Gate Set Tomography experiment on a superconducting quantum processor.   

SKQuant-Opt: Optimizers for Noisy Intermediate-Scale Quantum Devices

Classical optimizers play an important role in quantum computing, e.g., in the hybrid Variational Quantum Eigensolver (VQE) and Quantum Approximate Optimization algorithms. They are used in hyperparameter tuning, calibration, machine learning, etc. Unfortunately, most of the easily accessible optimizers do not handle noise well, leaving them below threshold for use with Noisy Intermediate-Scale Quantum (NISQ) devices. We present skquant-opt, part of scikit-quant.org, a set of optimizers tuned for the needs of NISQ. We have taken the state-of-the-art optimizers and tested them on a range of VQE applications and on hyperparameter tuning for optimization on D-Wave. Mesh methods, including hybrids that add local models, yield the best results. We present these results as well as guidance on use. Collected in skquant-opt, we provide the best optimizers in Python, the most used language in quantum computing, through the standard channels. The interfaces are made consistent with default parameters attuned to quantum computing problems, allowing for easy application and fast evaluation.