Processing Unit Scheduling Using Pseudo-Random Operation Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data processing techniques, such as neural network and graphics processing, face challenges in efficiently handling large amounts of data due to the complexity of managing diverse operations and the need for external memory transactions, particularly when executing operations in hardware accelerators like GPUs and NPUs.
Innovation Solution
A method and system for data processing that involves generating pseudo-random numbers based on scheduling-independent parameters to dynamically schedule operations, allowing for efficient conversion between higher and lower precision formats, and utilizing a centralized handling unit to manage transforms from operation space to section space, enabling simplified execution units and easy addition of new operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If operations are dynamically scheduled in hardware accelerators, then processing efficiency is improved, but the complexity of managing diverse operations increases
Solution Approach 1:
The patent uses pseudo-random number generation based on scheduling-independent parameters to dynamically schedule operations. By changing the parameters used for scheduling (using pseudo-random sequences derived from operation-specific parameters rather than complex scheduling algorithms), the system achieves efficient dynamic scheduling while reducing the complexity of managing diverse operations.
2Reliability
If external memory transactions are used for data processing, then data availability is improved, but the number of memory transactions increases
Solution Approach 1:
The patent segments the data processing workflow into distinct operational phases that can be executed with minimized memory access. By organizing operations into structured sequences with predictable memory access patterns, the system reduces the total number of external memory transactions while ensuring data availability when needed.
3Productivity
If precision conversion operations are performed, then computational efficiency is improved, but numerical precision is reduced
Solution Approach 1:
The patent employs dynamic precision management where operations can convert between higher and lower precision formats based on computational requirements. The system adaptively adjusts precision levels during execution, using higher precision when accuracy is critical and lower precision for intermediate calculations, thereby optimizing the balance between computational efficiency and numerical precision.
Data Source
AI summary
A method for executing a task using a processing unit, wherein the task comprises at least one operation. The method comprises obtaining, by a command unit of the processing unit, a pseudo-random number, and scheduling by the command unit, the at least one operation. The command unit generates at least one second pseudo-random number based on the pseudo-random number, and one or more scheduling-independent parameters relating to the operation. The scheduling-independent parameters are independent of the scheduling of the operation. The processing unit executes the at least one operation based on the at least one second pseudo-random number.