Masked Tensor AI Function Execution Without Explicit Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing environments face inefficiencies in performing tensor computations, particularly in deep learning, due to limitations in general-purpose processors and the need for hardware-specific programming that increases complexity and verification costs, and existing methods for masked artificial intelligence functions require performance-degrading operations.
Innovation Solution
A computer program product and method that allows for performing artificial intelligence functions on a subset of vector elements of an input tensor, simulating masked AI functions without explicit masking, thereby reducing processing time and resource usage by controlling the number of elements processed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If many instructions are executed on a general-purpose processor to perform AI workloads, then the processor can handle diverse computing tasks, but the performance of neural network operations is limited
Solution Approach 1:
The patent introduces a masked AI function execution mechanism that acts as an intermediary between the general-purpose processor and the AI workload. By using masking indicators to selectively enable or disable AI function execution, the system allows the processor to maintain versatility while optimizing performance for neural network operations through controlled instruction execution paths.
2Productivity
If a purpose-built hardware implementation is used for AI functions, then processing performance is optimized, but the program must be modified and recompiled for each hardware generation, increasing complexity and verification costs
Solution Approach 1:
The patent creates a universal instruction set that can be executed across different hardware generations without requiring recompilation. By defining AI function execution through a standardized interface with masking indicators, the same program can run on various processor architectures, eliminating the need for hardware-specific programming while maintaining optimized performance.
Solution Approach 2:
The system uses masking indicator parameters to control AI function execution without changing the underlying instruction architecture. By adjusting the mask indicator values rather than modifying the program code, the system adapts to different hardware configurations and workloads, reducing verification complexity while maintaining performance optimization.
3Ease of operation
If masked AI functions are implemented using existing methods, then AI function behavior can be controlled, but performance-degrading operations are required
Solution Approach 1:
The patent applies preliminary masking of instructions before execution, where the mask indicator is set in advance to define which AI functions should be executed. This preliminary configuration allows the processor to execute only the necessary instructions without requiring runtime performance-degrading operations to mask or disable functions, thereby maintaining high processing speed while achieving controlled AI function behavior.
Data Source
AI summary
Tensor processing includes obtaining an input tensor, the input tensor including a dimension of index size n, determining an element count, c, based on an indicator, the indicator specified by the instruction, and the element count specifying a number of vector elements on which to perform an artificial intelligence function, obtaining an input vector, of the input tensor, of size n, and performing the artificial intelligence function, the performing the artificial intelligence function including performing the artificial intelligence function on a first c number of elements of the input vector to provide a corresponding c number of elements of an output vector of index size n of an output tensor.