Single/double/quad precision floating-point data correlation circuit and operand acquisition method of a risc processor

Abstract

The invention provides a single / double / quad precision floating-point data related circuit and operand acquisition method of a RISC processor, which can effectively solve the problems of logic complexity and long delay, and can quickly determine floating-point single / double / quad precision The correlation, obtain the correct floating-point data source, and increase the frequency of the floating-point pipeline. By adopting 4 sets of hardware logic to carry out the correlation judgment and the processing of each part of data selection in parallel, the correlation judgment logic of floating-point data with different precision, and the complexity and long delay of data selection logic during correlation are eliminated. Compared with the logic of determining the correlation of each execution level according to the register number of the floating-point operand in the floating-point instruction and the precision of the floating-point operation, and selecting the floating-point result according to the relevant situation, the complexity is reduced and the logic delay is reduced. The circuit structure is clear, and the correlation determination preprocessing, correlation determination logic, and result selection logic can all be divided into pipeline levels, which is suitable for processors with different pipeline divisions.

Description

technical field [0001] The invention relates to the related design field of floating-point pipeline data, in particular to a single / double / quad precision floating-point data related circuit and an operand acquisition method of a RISC processor. Background technique [0002] At present, there have been many researches on the floating-point design of processors, but there are relatively few studies on floating-point pipelines. Since floating-point data includes single / double / quad precision, this means that single-precision floating-point 32-bit data may be related to single-, double-, and quad-precision floating-point data; double-precision floating-point 64-bit data may be related to single, double, Four-precision floating-point data is correlated, and the correlation between the upper 32-bit and lower 32-bit data in 64-bit data may be different; the quad-precision floating-point 128-bit data may be correlated with single, double, and quadruple-precision floating-point data, ...

AI Technical Summary

Problems solved by technology

[0003] Regarding single / double / quad precision floating-point data, there are many different related situations according to different precisions, and there are data related judgments based on various possible related situations, as well as the selection of operands, the problems of high logic complexity and long delay, and limitations frequency of the floating-point pipeline

The decision logic for each precision floating-point data is based solely on various possible related situations, and after determining the correlation, the selection logic of the correct data is more complicated, and the complex logic delay has become a bottleneck that limits the increase in the frequency of the floating-point pipeline.

At the same time, some processors use stall pipeline technology to determine that if data correlation occurs, they wait for the execution of the relevant floating-point instructions to complete, and then read the corresponding floating-point data from the floating-point register file. This method does not really reduce the floating-point data. At the same time, because the pipeline is paused, the throughput rate of the pipeline is reduced, and the performance of the floating-point pipeline is directly reduced.

Images