Floating-Point Multiply-Add Shift Prediction Circuit
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The long calculation delay caused by determining shift amounts in data processing operations, particularly in AI data processing involving floating-point numbers, leads to low data processing efficiency due to the large shiftable range and time-consuming shift processes.
Innovation Solution
A data processing method that involves reading floating-point numbers, performing multiplication, determining exponent differences, and aligning shifts based on a threshold, predicting normalization shifts, and performing normalization shifts synchronously to reduce calculation delays and improve efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional sequential processing of multiply-add operations is used, then calculation accuracy is maintained, but calculation delay increases and data processing efficiency decreases
Solution Approach 1:
The patent performs alignment shift operations in advance before the adder completes its calculation. Specifically, the alignment shift module performs the shift operation based on the exponent difference between the multiplication result and the third floating-point number, so that when the addition operation is completed, the normalization shift can be directly determined without waiting for intermediate calculation results. This preliminary action reduces the overall calculation delay and improves data processing efficiency.
2Measurement precision
If the full range of shift amounts is considered, then precision is maintained, but the time required to determine shift amounts increases
Solution Approach 1:
The patent segments the shift amount determination process into two distinct parts: alignment shift amount and normalization shift amount. The alignment shift amount is determined based on the exponent difference between the multiplication result and the third floating-point number, which can be calculated immediately. The normalization shift amount is determined after the addition operation based on the actual result. This segmentation allows the alignment shift to be performed in advance without waiting for the addition to complete, reducing the time to determine shift amounts while maintaining precision.
Data Source
AI summary
A method includes: performing multiplication on first and second floating-point numbers of a multiply-add operation to obtain a multiplication result; determining an exponent difference between exponents of the multiplication result and a third floating-point number of the multiply-add operation; performing alignment shift on the third floating-point number and/or the multiplication result in response to the exponent difference being less than or equal to a shift amount threshold, to obtain an aligned floating-point number and an alignment multiplication result; calculating an addition result between the aligned floating-point number and the alignment multiplication result by using an adder, and predicting a normalization shift amount for the addition result using a prediction encoder; and performing normalization shift on the addition result according to the predicted normalization shift amount, and obtaining an operation result of the multiply-add operation according to the obtained normalized result.