Relatore: Alex McCaskey, NVIDIA
As quantum computers continue to mature and progress, researchers will require robust and performant classical software architectures that provide a familiar toolchain for programming, compilation, emulation, and debugging. Chief among these tools will be those that enable rapid algorithmic prototyping via quantum circuit simulation. Advancing the state-of-the-art for quantum circuit simulation techniques is critical in the near-term for pertinent verification and validation workflows as researchers port quantum algorithms to physical quantum architectures. There is a unique opportunity to leverage the state-of-the-art with regards to classical heterogeneous computing, specifically multi-GPU compute systems. We present cuQuantum: an SDK for quantum circuit simulation that leverages NVIDIA GPU architectures via a low-level API that ultimately delegates to the CUDA GPU programming infrastructure. cuQuantum includes libraries to accelerate both state vector and tensor network simulation, and is already being adopted by leading simulation frameworks and integrated into the workflows of supercomputing centers. We will demonstrate recent results on NVIDIA GPU compute architectures showing world-class performance for both state vector and tensor network simulation on key quantum algorithmic primitives.