Offloading Server Automates GPU Code Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for offloading processing to GPUs are inefficient for unskilled users, requiring technical expertise and trial-and-error approaches to achieve good performance, often resulting in suboptimal CPU-GPU data exchange overhead and limited acceleration of applications.
Innovation Solution
An offloading server that analyzes application code, identifies loop statements, and uses explicit directives to optimize data transfers and parallel processing, employing a genetic algorithm to automatically select the most effective parallel processing patterns and reduce data transfer inefficiencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual programming and settings according to hardware (CUDA, OpenCL) are required to utilize GPU or FPGA, then processing performance can be improved, but the technical hurdle and complexity increase significantly
Solution Approach 1:
The system performs self-service by automatically analyzing application source code, identifying parallelizable processing sections, and generating optimized accelerator code without requiring user intervention. The offloading server autonomously completes tasks that would otherwise require expert programming knowledge in CUDA or OpenCL.
Solution Approach 2:
The offloading server acts as an intermediary between the application developer and the accelerator hardware. It translates high-level application code into optimized accelerator code, shielding users from the complexity of direct hardware programming while enabling efficient GPU or FPGA utilization.
2Productivity
If trial-and-error approaches are used to achieve good performance in CPU-GPU data exchange, then processing performance can be improved, but time consumption and complexity increase
Solution Approach 1:
The system performs preliminary analysis of the application source code to identify optimal data transfer points and parallelizable sections before actual execution. By pre-determining the best optimization strategy through static analysis, it eliminates the need for time-consuming trial-and-error runtime adjustments.
Solution Approach 2:
The system incorporates performance measurement functionality that provides feedback on actual accelerator performance, allowing automatic adjustment and optimization of data transfer strategies based on measured results rather than relying on manual trial-and-error approaches.
3Ease of operation
If automatic offloading of application processing to accelerator is implemented, then ease of use improves, but processing performance may be reduced without proper optimization
Solution Approach 1:
The system segments the application source code to identify specific parallelizable processing sections that are suitable for accelerator offloading. By carefully selecting and separating only the appropriate portions for acceleration while keeping other sections on the CPU, it maintains ease of automatic offloading while ensuring optimal processing performance.
Data Source
AI summary
An offloading server includes: a data transfer designation section configured to analyze reference relationships of variables used in loop statements in an application and designate, for data that can be transferred outside a loop, a data transfer using an explicit directive that explicitly specifies a data transfer outside the loop; a parallel processing designation section configured to identify loop statements in the application and specify a directive specifying application of parallel processing by an accelerator and perform compilation for each of the loop statements; and a parallel processing pattern creation section configured to exclude loop statements causing a compilation error from loop statements to be offloaded and create a plurality of parallel processing patterns each of which specifies whether to perform parallel processing for each of the loop statements not causing a compilation error.