The Key Technology Study For Super Precision Floating-Point Arithmetic

The floating point arithmetic is very important for high performance arithmetic. In order to satisfy the requirement of applications, some of the microprocessor and high performance vision card have the performance to deal with super precision arithmetic. But for the methods of how to realize super precision arithmetic, there is no more information.In this paper we mainly study for the key technology of super precision floating point arithmetic. The precision analysis of super precision arithmetic, super precision adder and super precision multiplying unit, are included in this paper. First we use the conception of errors in numerical analysis, design some new models, and ten discuss the error precision in super precision arithmetic. In a new point of view, we discuss the importance of study for super precision arithmetic. Second, in this paper we study the mainly technology of floating point arithmetic. In this part we give the principle of two data paths adder and three data paths adder, and the structure of super precision floating point adder. In the super precision floating point addition, significant digits addition and how to deal with leading zeros become more important for the reason of the significant digits went up. In order to deal with the significant digits addition we designed three inputs adder tree, and give a simple performance discuss of this method. For the leading zeros disposal, in this paper we studied the principle of leading zero detector and leading zero arbitral logic. Thirdly, in this paper we studied the main principle of how to design a multiplying unit. In this part it consists of part product produce and part product adder and how to rounding to the standard floating point. For the sake of save hard ware, in this paper we give a new method for rounding to the standard floating point. In order to reduce the number of the part products which take part in the add process, we give a method of how to combine some modules. And then control the number to be no more then twice of the significant digits portion. And then we give a simple experiment on our methods. For the last part we sum up our work, and give the main succulent work.