Реализация t-системы с открытой архитектурой для CUDA-устройств с поддержкой динамического параллелизма и для гибридных суперЭВМ на их основе

The paper outlines the basic principles of implementing the expansion of the open system T-system (OpenTS), which extends the T ++ programming paradigm to the hardware of modern special machines with CUDA technology support. A specialized version of the T-add-on implemented in the microcode of the OpenTS system is able to work autonomously inside the CUDA device, effectively turning it into a full-fledged intelligent T-node of a hybrid supercomputer. In combination with the support of dynamic parallelism with the latest CUDA-devices, it helps to significantly increase the percentage of utilization of graphics accelerators (GPU) without manual balancing of statically parallel blocks of the program. The universal logic of parallelization in the T-system is now able to generate and run lightweight countable granules, avoiding the losses that occur when the GPU interacts intensively with the processes of the basic cluster OS. The range of applied problems that are suitable for the new model of computation essentially replenishes the usual computing cores for the GPU, bringing arbitrary control logic to the level of special computers. As a demo example, the article deals with the problem of reversing cryptographic hash functions. The basic information related to applied applications of hash functions is given.