Tensor Matrix Multiplication With Dynamic Common-Dimension Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing environments face inefficiencies in tensor computations due to limitations in general-purpose processors and the need for hardware-specific programming, leading to performance bottlenecks and increased complexity.
Innovation Solution
A computer program product that enables dimension control for tensor matrix multiplication by allowing selection of common dimensions for input tensors, avoiding software transposition and optimizing processing through flexible dimension indicators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If many instructions are used on a general-purpose processor to perform tensor computations, then flexibility in processing is maintained, but processing speed and performance deteriorate
Solution Approach 1:
The patent segments the tensor computation process into distinct phases: dimension control indicator decoding, tensor selection based on controlled dimensions, matrix multiplication execution, and result generation. This segmentation allows the system to optimize each phase independently while maintaining overall flexibility in handling different tensor operations.
Solution Approach 2:
The patent introduces dynamic dimension control through the dimension control indicator, which allows the common dimension between input tensors to be dynamically specified rather than fixed. This dynamic approach enables the system to adapt to different computational scenarios while maintaining high processing speed through efficient instruction execution.
2Productivity
If purpose-built hardware implementation is used for tensor computations, then processing speed is improved, but programming complexity increases due to modification and recompilation requirements for each hardware generation
Solution Approach 1:
The patent creates a universal instruction format that can be executed across different hardware generations without modification. The dimension control indicator mechanism provides a standardized interface that works consistently across hardware versions, eliminating the need for program modification and recompilation while maintaining optimized processing speed.
Solution Approach 2:
The patent uses parameter changes in the dimension control indicator to adapt the computation to different hardware capabilities and tensor configurations. By encoding dimension control information in the instruction parameters, the system can optimize for specific hardware generations without requiring program changes, thus reducing programming complexity while maintaining processing speed.
3Ease of operation
If fixed dimension assumptions are made in conventional matrix multiplication methods, then processing simplicity is maintained, but adaptability to different tensor configurations deteriorates
Solution Approach 1:
The patent introduces the dimension control indicator as an intermediary between the instruction and the tensor operation. This intermediary carries dimension control information that mediates between the simple instruction format and the complex variable tensor configurations, allowing the system to maintain processing simplicity while achieving high adaptability to different tensor dimensions and configurations.
Data Source
AI summary
Dimension control in tensor multiplication includes obtaining first and second input tensors for matrix multiplication, obtaining a dimension control indicator that indicates a first dimension for the first input tensor to use as a common dimension for the matrix multiplication and indicates a second dimension for the second input tensor to use as the common dimension for the matrix multiplication, and performing the matrix multiplication to obtain one or more results, where performing the matrix multiplication includes selecting at least one vector of the first input tensor based on the first dimension indicated by the dimension control indicator and selecting at least one vector of the second input tensor based on the second dimension indicated by the dimension control indicator.