A Normalized Floating Point Data Screening Circuit

Abstract

The invention discloses a normalized floating-point data screening circuit. The circuit comprises a normalized floating-point data comparator, a rising edge monitoring circuit, an input circuit and an output circuit; a floating-point data stream Ain and an address stream Bin corresponding to the floating-point data stream Ain are input through the input circuit; a result R is output to the output circuit through the normalized floating-point data comparator; and obtained maximum / minimum data MData, an address MAddr corresponding to the data, an enable signal Men, a threshold comparison result ThresholdData and a current threshold comparison valid signal ThresholdEn are output. According to the circuit, a large amount of normalized floating-point data can be screened in a data stream mode; the data or the corresponding address meeting a specified condition can be given; and a screening apparatus is designed in an ALU, so that the floating-point data can be quickly screened, the design of a software program can be simplified, the executive energy consumption of a processor can be reduced, and the deficiency that a general data comparison instruction only can perform comparison on relatively few numbers (generally two numbers) can be made up for.

Description

【Technical Field】 [0001] The invention belongs to the technical field of digital circuits, and relates to a standardized floating-point data screening circuit. 【Background technique】 [0002] In digital signal processing chips, there is widespread processing of a large amount of floating-point data. In the processing of these floating-point data, in addition to the most basic addition, subtraction and multiplication, there is also a comparison of floating-point data. To compare two normalized floating-point numbers, first compare the sign bit, if the sign bit is the same, then compare the exponent size, if the exponents are also the same, then further compare the mantissa size, this is the best way to compare two normalized floating-point numbers The basic method. With the continuous expansion of the scale of processed data, it is often necessary to filter out data that meets a certain size relationship from a large number of data, and give the data value that meets the conditi...

AI Technical Summary

Problems solved by technology

In this way, in order to complete the screening of a set of data, it needs to occupy more register resources, and the code set formed by multiple instructions needs to occupy a large program memory space, and it needs to continuously execute value fetching, decoding, and execution. Waiting for operations, the code execution time is longer, and it will cause more energy consumption

Images