Runtime Library Patching for GPU Workload Profiling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Profiling operations for GPUs are labor-intensive and prone to errors due to manual code modifications required for instrumentation, slowing down development and reducing efficiency.
Innovation Solution
Automated runtime patching of code portions using patch libraries that insert profiling instrumentation without modifying developer-generated code, by replacing or modifying imported library segments with additional functionality for profiling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual code modifications are used to add profiling instrumentation, then profiling data can be obtained, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The patent creates a copy of the original code (the profiled application) and modifies only the copy by inserting profiling instrumentation through patching. This allows the original code to remain unchanged while still obtaining profiling data, eliminating the need for developers to manually modify their source code and reducing development time.
Solution Approach 2:
The patent introduces a patching mechanism as an intermediary between the original code and the profiling process. Instead of directly modifying the application code, the system applies patches that insert profiling instrumentation automatically, serving as a mediator that enables profiling without manual code changes.
2Measurement precision
If manual code modifications are used for profiling, then instrumentation can be added, but the process is prone to errors and requires developer skill
Solution Approach 1:
The patent implements self-service by allowing the profiling system to automatically identify and patch the necessary code sections without requiring developer intervention. The system autonomously determines which code portions need instrumentation and applies the appropriate patches, eliminating the need for developers to manually add profiling code and reducing the skill requirement.
Solution Approach 2:
The patent performs preliminary actions by pre-defining patch templates and instrumentation patterns that can be automatically applied. The system prepares and stores profiling instrumentation code in advance, allowing it to be automatically inserted into the application without requiring developers to manually write or configure the instrumentation code.
3Measurement precision
If manual code modifications are used for profiling, then profiling can be performed, but code changes must be reverted after profiling
Solution Approach 1:
The patent creates a separate copy of the application code that contains the profiling instrumentation, leaving the original code unchanged. This allows profiling to be performed on the copied version without permanently modifying the developer's source code, eliminating the need to revert changes and simplifying the workflow.
Solution Approach 2:
The patent implements dynamic patching where the instrumentation can be added and removed flexibly without modifying the original codebase. The patching mechanism allows the profiling instrumentation to be dynamically applied only when needed and easily removed afterward, providing dynamic control over the profiling process without permanent code changes.
4Productivity
If automated patching is used, then manual intervention is reduced, but library code must be modified
Solution Approach 1:
The patent applies local quality by modifying only specific portions of the library code (the imported modules that are actually called during execution) rather than the entire library. The patching is localized to only those functions and methods that are dynamically invoked, minimizing the scope of code modification while maintaining high profiling efficiency.
Solution Approach 2:
The patent uses dynamic patching that automatically identifies and modifies only the library segments that are actually called during the application's execution. This dynamic approach ensures that only the necessary portions of the library code are modified for profiling, rather than statically modifying the entire library, thereby reducing the overall complexity of code modification.
Data Source
AI summary
Approaches presented herein provide systems and methods for runtime patching of one or more code portions for an application undergoing a profiling operation. The application may include code portions that are imported, such as from libraries, that include a set of original functionality. Different code portions may be identified and then patched to include additional functionality, in addition to the original functionality, within the code portions that are to be imported at runtime. When the code portions are executed, the additional functionality may then be executed without modifying or changing the underlying developer-produced code.