CGRA Context Mask Loading for Faster Kernel Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing architectures face inefficiencies in data transfer between processors and memory, constraining performance and capacity, particularly in compute-intensive operations like neural networks and simulations, due to time and energy consumption.
Innovation Solution
Implementing a compute-near-memory (CNM) system with hybrid threading processors and coarse-grained reconfigurable array processors (CGRA) that utilize context masks to efficiently load only the necessary context state for kernel execution, reducing context loading time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional Von Neumann architecture with shared memory and bus is used, then system structure is simple, but data transfer between processors and memory consumes significant time and energy, constraining performance and capacity
Solution Approach 1:
The patent segments the monolithic processor architecture into multiple processing elements (PEs) organized in an array, with each PE capable of independent operation. This segmentation allows parallel processing of multiple data elements simultaneously, improving compute performance while reducing the time each element needs to wait for data transfer, as operations can proceed in parallel across the array.
Solution Approach 2:
The patent introduces a reconfigurable interconnection network as an intermediary between the processing elements and memory systems. This intermediary enables direct data routing and transfer between PEs and memory without requiring all data to traverse a single shared bus, thereby reducing data transfer time and enabling parallel access paths.
2Productivity
If full context state is loaded for kernel execution, then context switching is complete and reliable, but context loading time increases
Solution Approach 1:
The patent extracts and loads only the necessary portions of context state required for kernel execution rather than loading the entire context state. By identifying and loading only the relevant context elements needed for the specific kernel being executed, the system reduces context loading time while maintaining complete and reliable context switching for the active kernel.
Solution Approach 2:
The patent applies partial action by loading a subset of the full context state - specifically, only the portions needed for the current kernel execution. This partial loading approach avoids the overhead of loading complete context state while ensuring all necessary information is present for correct kernel operation, thereby improving execution efficiency without sacrificing reliability.
Data Source
AI summary
Devices and techniques for loading contexts in a coarse-grained reconfigurable array processor are described herein. A system or apparatus may include context load circuitry operable to load context for a coarse-grained reconfigurable array processor, where the context load circuitry is configured to: (a) receive a kernel identifier; (b) access a first registry to obtain a context mask base address; (c) determine a context mask address from the context mask base address and the kernel identifier; (d) access a second registry to obtain a context state base address; (e) determine a context state address from the context state base address and the kernel identifier; (f) use a context mask at the context mask address to determine corresponding active context state; and (g) load the corresponding active context state into the coarse-grained reconfigurable array processor.