Effectiveness of low/adaptive precision with ICCG method

Low precision computation is widely used in the field of deep learning. In contrast, in the field of numerical computation, the utilization of the low precision is limited due to concerns about accuracy. In this study, we apply and evaluate the effectiveness of the low-precision to a conjugate gradient (ICCG) iterative method with incomplete Cholesky precondition, which requires particularly high arithmetic accuracy. In addition, we also consider adaptive precisions, which have been studied in numerical simulation in recent years.