Floating-Point Alignment Sticky Bit Computation via Segmented Pre-computation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional methods for computing the alignment sticky bit in floating-point operations fail to balance speed and implementation complexity, especially in adding floating-point support to integer processor pipelines.

Innovation Solution

A method that computes the alignment sticky bit by pre-computing a sticky bit and using a shifter sticky OR, or computing a plurality of sticky group ORs, allowing for parallel computation and improved efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional techniques are used to calculate the sticky bit, then the calculation can be performed, but the speed and implementation complexity are not balanced

Engineering Contradiction:

Improvespeed of sticky bit calculationVSAvoidimplementation complexity of sticky bit calculation

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The sticky bit calculation is divided into multiple independent stages: pre-computing sticky groups from blocks of significand bits, computing sticky group ORs, and finally computing the alignment sticky bit. This segmentation allows parallel execution of operations and reduces the complexity of any single stage while maintaining high speed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Sticky groups are pre-computed from blocks of significand bits before the final alignment sticky bit calculation is needed. This preliminary action enables the final computation to proceed faster by having intermediate results ready, and it distributes the work over time to reduce instantaneous complexity.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If more bits are stored during alignment, then precision is maintained, but the complexity of computing the sticky bit increases

Engineering Contradiction:

Improveprecision of floating-point operationVSAvoidcomplexity of sticky bit computation

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The significand is divided into multiple sticky groups of fixed size, and each group is processed independently to produce a sticky group bit. This segmentation reduces the complexity of processing individual groups while maintaining the precision benefits of tracking all shifted-out bits through the hierarchical OR structure.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The problem is transformed from computing a single sticky bit directly to computing multiple sticky group bits in parallel across different dimensions (bit groups), then combining them through hierarchical OR operations. This dimensional transformation reduces computational complexity while preserving precision.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS8543632B2Method and system for computing alignment sticky bit in floating-point operations

Publication Date: 2013.09.24 STMICROELECTRONICS INT NV

AI summary

A method for computing the alignment sticky bit in floating-point operations is provided. The method includes computing a pre-computed sticky bit. A significand is aligned based on an alignment counter. A shifter sticky OR is computed. The alignment sticky bit is computed based on the pre-computed sticky bit by ORing the pre-computed sticky bit and the shifter sticky OR when the alignment counter comprises a value greater than or equal to a predefined value.