Research On Key Technologies Of Custom Processor And Compiler For The Orbit Prediction Of The Spatial Object

The orbit prediction of spatial object is one technology, which make use of the mathematical model to predict the track of the spatial object within a period of time in the future based on the motion characteristic of the spatial object. The orbit prediction is the basis of the applications that are real-time monitoring, collision warning, and weapon attacks, which is widely used in the fields, such as aerospace, military fields, and so on.The orbit prediction is characterized by dense data dependence, computation complexity, huge amount of calculation, and the quantity of the spatial object is constantly increasing. All of these put forward higher requirement to the system's data processing capacity. It is impossible for a single processor computer to accomplish the orbit prediction of all the spatial objects, and the system of cluster or giant computer has many problems, such as high power consumption, complexity control, inability to custom, low efficiency, and so on. Reconfigurable devices such as Field Programmable Gate Arrays are ideal candidates for the hardware-based acceleration of algorithms, because of their high density, high speed, low power consumption and flexibility. It is a new efficient method to promote the performance of a particular application through making use of Field Programmable Gate Arrays chips to construct the custom processor.Based on the analysis of the orbit prediction program, which bases on the Simple General/Deep-space Perturbation Version4 model, this paper adopts a series of optimizing technologies, such as exploiting multi-hierarchy parallelism, resource share, predicated execution, and so on, to design and realize a custom processor with the very long instruction word structure, which is oriented to the orbit prediction. The experiment result shows that comparing to the software program running on the platform of QuadCore AMD Phenom X4 9650 CPU the custom processor integrated with eight processing elements can achieve the speedup which is more than 8X.It is very complex and easy to go wrong that translating the source code of the orbit prediction program into the target code of the custom processor through manual encoding. In order to solve the problem, according to the characteristic of the custom processor, this paper designs and realizes a custom compiler. Through adopting a series of optimizing strategies, such as instruction schedule based on the cycle and operation, straight, heuristic memory allocation, predicated execution, and so on, the compiler can realize the translation efficiently and automatically.In order to make the compiler support the custom processors with different structures, this paper puts forward a template of the target machine architecture through analyzing the optimizing process, and designs a method to describe the template based on the thinking of the oriented-object. In this way, we can separate the design and realization of the compiler from the details of the target machine architecture. Finally this paper takes the instruction schedule for an example to demonstrate the process of analyzing the template, and realize the retargetable compiler.