AO development platform

The simulation of an AO system involves multiple physics from atmospheric turbulence models to tomographic reconstruction to control theory. Due to the stochastic nature of the turbulence, Monte Carlo simulations provide the most realistic results.

Full length E-ELT simulations are compute-intensive applications and as such good candidates for considering the use of hardware accelerators like manycore processors. Among those accelerators, the CUDA architecture was designed to provide graphics processors (GPUs) equipped with High Performance Computing (HPC) compatible features (e.g. IEEE754-like floating point unit, memory model) offering a tremendous performance potential at a very low power.

The end product of the AO development platform of COMPASS is a unified and optimized computing framework (software and hardware), will address several major needs for research in AO as it provides :

• An efficient computing environment to run large scale AO systems simulations in almostreal-time in order to test the design of the E-ELT instrumentation and prepare the scientific return of the foreseen instruments
• A real-time core, integrated in the simulation environment, fully expandable and capable of driving a real system at the appropriate speed using various command /control algorithms depending on the system (SCAO, MCAO, MOAO)
• A high bandwidth low latency interface between such a real-time core and a standard interconnect such as 40 GbE
• A pathfinder for accelerator-based AO control under the form of a prototype for a low latency image acquisition and processing system including the real-time core and its interface to a camera

Adaptive Optics simulation tool

YoGA AO is based on the CUDA C toolkit and an original binding to Yorick an open source interpreted language providing the user with a comprehensive interface. It includes a multiple layers turbulence model, a Shack-Hartmann wavefront sensor
model for natural (NGS) and laser guide stars (LGS), a simple real-time computer (RTC) module and a real-istic deformable mirror model. It is highly configurable and allows to simulate a variety of systems from single conjugate NGS AO to multiple guide stars LGS AO.

The original binding between Yorick and CUDA (YoGA) on which it relies allows to manipulate persistent objects on the GPU from within a Yorick session and thus launch intensive computations on these objects through an interpreted environment.

The simulation code is open and distributed free of charge under a GPL licence. Installation instructions and various information are available on the wiki

Pathfinder for accelerator-based real-time control