Vector Processor Data Type Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vector processors face inefficiencies in using operation units due to high Byte/Flop (B/F) values, leading to wasted resources and poor performance in executing operations, especially when handling multiple elements as separate data types.
Innovation Solution
A vector processor configuration that includes a temporary storage device, a data type determining part, and an output destination deciding part, allowing elements to be collectively handled as one element based on the instruction, optimizing data type and output destination decisions for efficient operation unit utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If elements are handled as separate data types in traditional vector processors, then data type precision is maintained, but operation unit utilization efficiency deteriorates due to high B/F values
Solution Approach 1:
The patent merges multiple elements of the same data type into a single data type determination result. The data type determining part determines one data type for multiple elements collectively, and the output destination deciding part decides a common output destination for these elements. This merging approach reduces the number of separate data type determinations and output destination decisions, thereby improving operation unit utilization efficiency while maintaining data type precision through collective handling.
Solution Approach 2:
The patent implements multi-functionality in the data type determining part and output destination deciding part. These parts can handle both individual element determination and collective multi-element determination. When multiple elements share the same data type, the system universally applies a single determination result to all elements, enabling the same hardware components to efficiently process varying numbers of elements without requiring separate processing paths for each element.
2Reliability
If multiple memory operations are performed for each operation, then data accuracy is ensured, but Byte/Flop ratio increases leading to resource waste
Solution Approach 1:
The patent merges the handling of multiple elements into a single operation flow. By determining one data type for multiple elements and deciding one output destination for these elements, the system reduces the number of separate memory operations required. This merging approach ensures data accuracy through consistent processing while reducing the Byte/Flop ratio by minimizing redundant memory access operations that would otherwise be required for each individual element.
3Reliability
If traditional vector processor architecture is used, then operational correctness is maintained, but performance is poor due to wasted operation unit resources
Solution Approach 1:
The patent introduces dynamic adaptability to the vector processor architecture. The data type determining part and output destination deciding part can dynamically adjust their operation based on the number of elements and their data types. When multiple elements share the same data type, the system dynamically consolidates the determination process, allowing the architecture to adaptively optimize resource utilization while maintaining operational correctness through consistent processing rules.
Solution Approach 2:
The patent changes the processing parameter from individual element-level determination to collective multi-element determination. By changing the scope of data type determination and output destination decision from single-element to multi-element basis, the system improves performance through better resource utilization while maintaining operational correctness through the consistent application of determination rules across all elements.
Data Source
AI summary
A vector processor includes: a temporary storage device configured to retain a plurality of elements representing data used at the time of performing an operation appropriate for an instruction; a data type determining part configured to determine what data type the elements retained by the temporary storage device are to be handled as among predetermined data types, in accordance with the instruction; and an output destination deciding part configured to decide an output destination of each of the elements stored by the temporary storage device, based on the result of determination by the data type determining part. The vector processor is configured to output each of the elements to the output destination decided by the output destination deciding part, thereby performing the operation.