Zone-Based Floating-Point Alignment Circuit for Reduced Hardware Use
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing floating point number operations in high-performance computing systems require efficient alignment of mantissas and exponents to facilitate accurate addition and subtraction, which current methods often consume significant hardware resources and energy.
Innovation Solution
A data processing circuit that includes a dual-mode adder, max finder circuit, zone detector circuit, and alignment circuit to align mantissas by classifying exponents into zones, reducing the need for full-bit comparisons and minimizing hardware and energy usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional full-bit exponent comparison and mantissa alignment methods are used, then mantissa alignment precision is improved, but hardware resource consumption and energy usage increase
Solution Approach 1:
The patent divides the exponent comparison process into two stages: first comparing the most significant bits (MSBs) to determine zone classification, then using this zone information to guide the alignment process. This segmentation allows the system to achieve precise alignment while reducing the computational burden on hardware resources by not requiring full-bit comparisons throughout the entire process.
Solution Approach 2:
The patent performs preliminary zone classification based on MSB comparison before completing the full mantissa alignment process. By pre-determining the zone based on exponent magnitude, the system can optimize subsequent alignment operations, reducing the need for exhaustive full-bit comparisons and thereby lowering hardware resource consumption while maintaining alignment precision.
2Measurement precision
If traditional full-bit exponent comparison and mantissa alignment methods are used, then mantissa alignment precision is improved, but energy consumption increases
Solution Approach 1:
The patent segments the alignment process into zone classification (using MSBs) and refined alignment (using zone information). This segmentation reduces the total number of bit comparisons required, thereby decreasing energy consumption while preserving alignment precision through the two-stage approach.
Solution Approach 2:
By performing preliminary zone classification using only the most significant bits of the exponent, the system reduces the energy required for comparison operations. This preliminary action guides subsequent alignment steps, avoiding the need for energy-intensive full-bit comparisons while maintaining the precision required for accurate floating-point operations.
3Device complexity
If zone-based alignment with limited bit comparison is used, then hardware resource consumption is reduced, but alignment accuracy may deteriorate
Solution Approach 1:
The patent uses preliminary zone classification based on MSB comparison to establish a framework for alignment. This preliminary action provides sufficient information to determine the relative positions of mantissas, enabling accurate alignment without requiring full-bit comparisons. The zone information serves as a guide that maintains alignment accuracy while reducing hardware requirements.
Solution Approach 2:
The patent applies different levels of comparison precision to different parts of the exponent-mantissa structure. By focusing full-bit comparison only where necessary (in the refined alignment stage guided by zone classification) and using MSB comparison for zone determination, the system achieves accurate alignment with reduced hardware resource consumption, applying local quality to optimize both accuracy and efficiency.
Data Source
AI summary
A circuit for data processing is provided. The circuit comprises a dual-mode adder, a max finder circuit, a zone detector circuit and an alignment circuit. The dual-mode adder generates products between first exponents of first floating point numbers and second exponents of second floating point numbers. The max finder circuit finds a maximum among first portions of the products. The zone detector circuit classify the first portions into zones by comparing the first portions and the maximum. The alignment circuit align first mantissas of the first floating point numbers according to the zones and second portions of the products to generate aligned mantissas for a floating point number operation.