In recent years, a new tomographic inversion method based on the Maximum Likelihood (ML) approach has been adapted to JET bolometry. Apart from its accuracy and reliability, the key advantage is its ability to provide reliable estimates of the uncertainties in the reconstructions. The original algorithm was implemented and validated using the MATLAB software tool. This work presents the accelerated version of the algorithm implemented using a compatible ITER fast controller platform with the Ubuntu 18.04 or the ITER Codac Core System distributions (6.1.2). The algorithm has been implemented in C++ using the open-source libraries: ArrayFire, ALGLIB, and MATIO. These libraries simplify the management of specific hardware accelerators such as GPUs and increase performance. The speed-up factor obtained is approximately 10 times. The work presents the methodology followed, the results obtained, and the advantages and drawbacks of implementation.

Image-based diagnostics are key for fusion experiments. The operating conditions at ITER and the future machines require changing the role of such systems from monitoring and archiving for offline post-processing to real-time processing. One of the roles of such systems is machine protection. A relevant application of vision diagnostics is the wall and divertor temperature monitoring and hot spot detection. However, algorithms for hot spot detection are computationally costly. To achieve real-time performance at the required time resolution for all these experiments, evaluating and validating the newest technologies is vital. This work applies heterogeneous computing techniques based on the OpenCL standard to the real-time hot spot detection problem and obtains performance values in an MTCA platform. OpenCL reduces the development time, improves portability, and simplify the evaluation and validation of each part of the algorithm to find the best-suited device in the heterogeneous system. The proposed solution enables balancing the computational load between an FPGA and a GPU. The algorithm has been adapted and optimized, taking profit on the particularities of each platform.

No abstract

No abstract